So I screen-shared my Jira board.

I dragged a card from Backlog to In Progress.

Within about two minutes, a podcast outline hit my Slack with real stories from my past conversations already tagged in… relevant ones, not random ones. The agent had searched my Supabase vector database, found the stories that matched the topic, and built the outline around them.

The Lindy team was quiet for a second. Then: Wait. Can you show that again?

Yeah. That is the one.

The problem content creation actually has

Most people think the bottleneck is writing. It is not. The bottleneck is starting.

Before I built this system, I would sit down to record a podcast and spend 20-30 minutes organizing my thoughts. Looking for the story I told three months ago that was perfect for this topic. Flipping through notes.

Not terrible. But I record 2-3 times a week. That friction compounds.

How the system actually works

Three pieces.

The Kanban board (Jira). Each quarter I plan my content topics and drop them into Jira as cards. When I am ready to create, I drag the card to In Progress. That drag is the trigger.

The webhook to Lindy. Jira fires a webhook the moment a card changes status. Lindy is listening. The agent receives the card data and that context becomes the input for what it builds.

The Supabase story database. Over the past couple years I have been running a transcript-first system where every meeting, coaching call, podcast recording, and workshop gets transcribed and stored. I extracted the stories from those transcripts and stored them in Supabase with vector embeddings.

So when Lindy gets the topic, it runs a semantic search against that database. Finds the 3-5 most relevant stories. Tags them into the outline.

Total time from drag to outline: a few minutes.

Why this works better than just prompting an LLM

A generic outline is fine. But generic outlines produce generic content.

When the outline already has a specific story from a real client call, the recording goes somewhere real. The stories are mine. The AI just surfaced them at the right moment.

Your past conversations are a goldmine. The problem is not that the material is not there. The problem is accessing it when you need it.

The bigger thing I noticed

Once you have a system synthesizing across your transcripts, you start seeing patterns you would never catch otherwise.

I added a companion workflow that runs every Friday. It reads the transcripts from all my meetings that week, sometimes 15 to 20 of them, and gives me a summary of recurring themes and patterns.

Reading one meeting in isolation, you cannot see it. But when an agent reads 20 meetings and tells you four different clients mentioned confusion around the same thing… that is actually useful intelligence. Not just notes. Insight.

That took maybe two hours to build on top of what was already there.

The 80-20 principle for agent building

Does this task happen daily or weekly? If yes, build it now. Is this a quarterly or one-off thing? Wait.

The math on compounding is brutal in your favor when something runs every week. A workflow I built in one weekend has now run over a hundred times. The ROI cleared in the first two weeks.

That is the first agent most people should build: whatever their version of I do this every week and the prep is annoying is.

How to build something like this

You do not need Jira or Supabase.

• A Google Sheet can replace the Kanban board. Add a status column. Use Zapier or Make to watch for status changes.
• A simple Airtable base with tagged notes can replace a vector database.
• The Lindy agent can be a Make scenario or a custom GPT with a retrieval plugin.

The core pattern: trigger on status change, look up relevant context, assemble structured output.

What changed after building this

The biggest change is not speed. It is energy.

When I sit down to record, I am not in logistics mode. I am in thinking mode from the start. The outline is there. The stories are there. I just bring the energy and the perspective.

Build the system once. Let it do the boring part. Show up for the part only you can do.

Want to build something like this? The best starting point is my Productivity Academy where we walk through agent building from scratch with real workflows.

When I show email automation, people nod. Calendar management, they're interested. Research tools, they're following along.

Then I show the meeting prep agent.

Something shifts in the room.

People start leaning forward. They start asking questions. And by the end of the demo, everybody wants to know: “Okay, how do I get this?”

It's been true in every single workshop I've run.

So let me tell you what it does, why it works, and how to think about building something like it for yourself.

What the Meeting Prep Agent Actually Does

Here's the core workflow.

An hour before each of your meetings, the agent wakes up. It looks at your calendar, grabs the attendees, and starts researching. It checks LinkedIn. It checks their social media. It scans your email history with that person.

Then it writes you a two-page brief: who they are, what you've discussed before, any relevant context, and suggested talking points.

You read it in 2-3 minutes. You walk into the meeting knowing exactly what you need to know.

Total setup time? Maybe 30 seconds of actual prep on your end. Because the agent already did it.

Compare that to what most people actually do: open a new tab, Google the person's name, scroll their LinkedIn, try to remember what you talked about last time, maybe dig through email for the thread. That's 10-15 minutes per meeting, every time.

If you have 6 or 7 meetings in a day, you're spending close to two hours just on pre-meeting admin. And the thing is, it doesn't feel like a problem. It feels like normal. You've always done it this way.

Why This One Lands When Other Automations Don't

I have a concept I use when helping clients build agents. I call it the 80-20 rule for automation.

The idea is simple: automate the things you do every day or every week first. Not the impressive rare stuff. Not the complex one-off project. The boring repetitive daily loop.

That's where compounding ROI lives.

Meeting prep is daily. For most knowledge workers, it happens multiple times a day. The time isn't all in one place, so it doesn't feel like a problem. But add it up… and it becomes one of the biggest drains on your day.

That's exactly why this agent creates such an immediate reaction when people see it. They calculate it in real time. “I have 8 meetings tomorrow. That's what, an hour and a half I just got back?”

Yeah. Pretty much.

A Real Example: 20 Hours a Week, Down to 15 Minutes

I was working with a client who runs a membership club in Austin. Packed schedule. He was spending something like 20 hours a week on meeting prep and follow-up combined.

We built him a daily briefing system. Each morning, it pulls from his calendar, looks up everyone he's meeting that day, and combines that with the email history and context from his CRM.

He reads it over coffee. 15 minutes. For the whole day.

That's not a small improvement. That's his entire week of prep, automated.

The interesting part: after he saw it working, he wanted to roll out the same system for everyone on his team. Because at that point the ROI is obvious. It's not “AI feels exciting.” It's “this thing just saved me two hours this morning.”

The Tool Selection Question

Here's something I get asked a lot after demos: “Do I build this in ChatGPT or in Lindy?”

The answer depends on what you need it to do.

If you want the agent to actually connect to your calendar, scan your email inbox, and pull LinkedIn data… you need Lindy (or a similar workflow automation tool). That's because you need to connect to external systems. Calendar APIs. Email. CRM.

ChatGPT is great for pure text work. If you want to take a bunch of notes and ask AI to organize them into a brief, ChatGPT is faster to iterate with. But it doesn't natively connect to your Gmail, your calendar, or your contact database.

I think of it this way: use ChatGPT for tasks where the information is already in front of you. Use Lindy when you need the AI to go out and gather the information for you.

Meeting prep requires gathering. So Lindy wins here.

The broader principle is what I call being multi-tool native. The best AI users aren't loyal to one platform. They route work to whichever tool fits the job. ChatGPT for content drafts. Gemini for visual tasks or multi-step research. Lindy for recurring automations that touch external systems.

How to Start Building This

You don't need to build the whole thing at once.

The simplest version: create a Lindy agent that takes a contact's name and runs a Google search plus LinkedIn lookup, then summarizes what it finds.

That's step one. A manual research helper you invoke before each meeting.

Once that's working, you add the calendar trigger. Now it runs automatically an hour before every meeting.

Then you add the email scan. Now it's pulling your actual conversation history.

Each step builds on the last. And each step alone is already useful.

That's the pattern I've seen work with every client. Don't design the full system first. Build the smallest useful version, confirm it actually saves time, then expand.

Most people over plan the agent they want. They want the full Digital Chief of Staff on day one.

Start with meeting prep. That's the thing everyone needs, everyone does manually, and takes maybe an afternoon to get working.

After that, adding features becomes incremental. You're not starting over. You're just adding to something that already works.

Want to see how this works in a live environment? I run one-day AI workshops in Austin where we build agents like this from scratch. Get in touch if you're interested.

One of the guys in the group had spent about two hours trying to build a calendar agent in Lindy. He was frustrated. The agent kept scheduling meetings he didn't want, missing the ones he did, and producing calendar summaries that weren't useful. He was ready to write off the whole thing.

I asked him: “What exactly did you tell it to do?”

He pulled up his prompt. It said something like: “Help me manage my calendar.”

That's the problem. That's almost always the problem.

The Doorman Analogy

Here's how I explained it to him, and to every workshop group since.

Think of your calendar agent as a doorman. Not just someone standing at an entrance. A really good doorman — the kind you find at a hotel where everyone knows your name. One who knows which guests are expected, who gets waved through, who waits in the lobby, and who gets politely turned away. And one who keeps you in the loop without you having to ask.

That's what a calendar agent should be.

It should protect your time. Filter what gets scheduled. Brief you before your day starts. And flag things that don't belong.

When I use that frame in workshops, something clicks for people. They stop thinking about the agent as a tool to configure… and start thinking about it as a role to fill.

And like any role you're hiring for, it needs a clear job description.

Vague Brief = Vague Output

I had a coaching session last year where a guy was stuck on a similar thing — couldn't get his AI to give him useful results on anything. He kept blaming the model.

When I pressed him, I realized the issue: he couldn't clearly explain what he wanted. Not to me, not to the AI.

I quoted something I come back to a lot: defining the problem is half the battle. And it's harder than most people expect.

The same applies to building agents. If you can't explain what good looks like, neither can the AI.

This is what I mean by the OCE prompt formula. Before you write a single line of your agent's instructions, get clear on three things: the Outcome you want, the Context the agent is working in, and your Expectations for format, timing, and behavior. Most people skip straight to expectations without ever nailing the outcome.

What Goes in a Calendar Agent Brief

So what does a good calendar agent brief actually include? Here's what I give mine, and what I walk workshop participants through:

1. What it can auto-accept

Be specific. “Accept any 30-minute coffee meeting from someone I've already met” is good. “Accept meetings that seem important” is not. The agent needs rules, not instincts.

2. What requires your review

Anything over 90 minutes. New contacts. Meetings that conflict with blocks you've protected. Write these out explicitly.

3. What to decline automatically

For me: same-day meeting requests, anything with vague agendas, calls scheduled for Friday afternoons. Your list will be different. Write it down.

4. A daily briefing format

This is the piece most people miss. A good calendar agent doesn't just execute — it keeps you posted. Mine sends me a morning summary that includes what's on the schedule, any changes from yesterday, and what's coming up in the next 48 hours. Two sentences is enough.

5. Edge case escalation

When in doubt, what should the agent do? Flag it for you? Decline with an apology? Hold the time pending your response? Define this. An agent without escalation rules makes its own decisions… and they're usually not the ones you'd make.

Meta Prompting: Let the AI Write the Brief for You

Here's a trick I showed in the same November workshop that got a lot of “wait, what” reactions.

You don't have to write the calendar agent prompt yourself. You can ask Claude or ChatGPT to write it for you.

I call this meta prompting. You describe your situation to the AI — the kind of schedule you keep, the types of meetings you take, your working hours, your scheduling rules — and then you ask it to write the agent prompt. A proper one, with role, goal, behavior instructions, examples of success, and a definition of done.

The prompts it generates are almost always better than what people write themselves. More complete. More specific. And they include components that people forget — like handling ambiguous requests or what to do when the agent encounters a calendar conflict.

One workshop participant tried this in real time. He spent maybe 10 minutes describing his work situation to Claude, asked it to draft a calendar agent prompt, and then copied the output into Lindy. His agent started working correctly almost immediately.

The insight is simple: if you're not sure how to write a brief, describe what you want to a smarter system and let it structure the brief for you. Then use that output as your agent's instructions.

The Logo Designer Test

There's another way I check whether an agent brief is good enough.

I ask: would you give this brief to a logo designer and expect a good logo?

If a client walked into a design studio and said “make me something modern, clean, and premium” — every designer in the building would groan. That's not a brief. That's a mood board. The designer needs to know the dimensions, the use cases, the colors to avoid, and what the client will actually do with the logo.

“Help me manage my calendar” is the AI equivalent of “make me something modern, clean, and premium.”

A good brief gives the agent a clear definition of done. Something it can check itself against. “The meeting is confirmed only when both parties have received a calendar invite and the meeting includes a Zoom link” is a definition of done. “The meeting is scheduled appropriately” is not.

When your agent has a clear definition of done, it stops guessing. And that's when it starts actually being useful.

The 15-Cent Proof

One of my clients had a scheduling workflow that took 30 to 60 minutes of back-and-forth every week. Coordinating calendars with multiple people, making sure his preferences were respected, chasing confirmations.

After we built a calendar agent with a proper brief — using all of the components above — that same workflow ran for 15 cents in API credits. One pass. The agent scheduled the right things, declined the wrong ones, and sent a summary he could read in 45 seconds.

That's not a feature of the technology. That's what happens when you give an agent a real job description instead of a vague hope.

Where to Start

If you want to try this, here's the simplest path.

Open a blank chat with Claude or ChatGPT. Describe your work situation: what kind of meetings you take, when you prefer to work, who you work with regularly, what your scheduling headaches are. Then ask it to write a calendar agent prompt using best practices for role, goal, context, samples of success, and definition of done.

Read what it produces. Edit anything that doesn't match your actual situation. Then copy it into your calendar agent tool of choice.

That's it. The whole thing takes less than 20 minutes.

The hard part isn't the tool. It was never the tool. It's giving the AI a job description clear enough that it can actually do the job.

If you want to see this built live, my one-day AI workshops in Austin walk through calendar agents, email agents, and meeting prep agents from scratch. Every workshop is hands-on. No slides, no theory — just building. Check the schedule at asianefficiency.com.

I'd drive out to the Texas Hill Country, find a cabin for the weekend, and spend two or three days in near-total isolation just to figure out what I wanted to focus on for the next 90 days. No distractions. No calls. Just me, my notebook, and a lot of walks.

It felt necessary. Like the clarity would only come if I physically escaped everything.

That was the old way.

Last December, I ran the same process in under 2 hours. At my desk. And honestly? The output was better.

Here's what changed.

AI Doesn't replace thinking. It Accelerates It.

The big misconception people have about using AI for goal setting is that the AI does the thinking for you. It doesn't. You still have to do the hard work of deciding what matters.

But what AI can do is get you to the hard questions faster.

When I used to go on those cabin retreats, I'd spend the first day just clearing my head. Writing down random thoughts. Dumping everything that had been bouncing around in my brain. Only on day two would I start making sense of it.

With AI, that initial dump-and-clarify phase takes about 20 minutes.

I open a session, share what's been on my mind, and within a few exchanges I'm already at the interesting questions — not still sorting through noise.

The Flip the Script Prompt

Most people use AI the same way they use a search engine. They ask it a question, it answers. That's fine for some things, but for goal setting it's backwards.

When I don't know exactly what I want or how to prioritize competing goals, I flip the script.

Instead of telling the AI what to do, I say something like: “Here's a rough list of goals I'm considering. I want you to help me prioritize them. Ask me what you need to know.”

And then it interviews me.

It asks about my timeline, what matters most this quarter, where I'm already making progress, what keeps getting pushed to the back burner. The questions it asks are often better than the ones I'd think to ask myself.

This works because the bottleneck in goal setting isn't usually information — it's clarity. You already know what's important. You just haven't articulated it yet. The back-and-forth of being interviewed surfaces things you wouldn't have written down on your own.

I've used variations of this with clients in workshops and they consistently say the same thing: “I didn't expect to learn something new about my own goals from that conversation.”

The Question Most People Skip

Here's the prompt I've found most useful in quarterly planning.

It's not “what should I focus on?” Most people ask that. The answers tend to be obvious.

The better question is: “Based on everything you know about me and my current situation, which of these goals should I NOT pursue — and why?”

That one lands differently.

When you ask AI to argue against your goals, it forces the same kind of hard thinking that used to take me three days in a cabin. It brings up the opportunity costs, the conflicts between competing priorities, the goals that only feel urgent because they're exciting right now.

And then you're left with what actually matters.

The subtraction philosophy is something I've come back to repeatedly at Asian Efficiency. It's easy to add more. Taking things off is harder and usually more valuable. This prompt is a fast path to doing exactly that.

How This Fits Into Quarterly Planning

I still believe in quarterly planning. The 12 Week Year approach — picking one meaningful goal and building everything around it — is one of the frameworks we've taught for years. The compressed 90-day cycle creates urgency in a way that annual planning just doesn't.

But the front end of that process, the part where you figure out what the right goal even is, used to be where I got stuck. Too many options. Too much noise. That's where the cabin retreats came in.

Now I use AI to compress that front end. Get clear on what matters, eliminate what doesn't, and show up to the 12-week sprint with confidence that I'm working on the right thing.

The retreat was never about the cabin. It was about having enough uninterrupted time to think clearly. AI gives me that… on demand, in a fraction of the time.

What to Try Before Your Next Planning Session

If you're heading into Q2 planning — or any quarterly review — try this sequence before your next session:

1. Write down every goal or project that's been on your mind. Don't filter. Just get it out.
2. Share the list with ChatGPT or Claude and say: “Ask me what you need to know to help me prioritize these.”
3. Answer its questions honestly. Let it push back.
4. Then ask: “Which of these should I NOT pursue right now and why?”
5. Take what's left. That's your quarter.

This works whether you're planning your business, your personal goals, or a mix of both.

The cabin was good. But you don't need the cabin.

Want to go deeper on AI-powered goal setting and quarterly planning? I cover this in detail in my Productivity Academy workshops and one-day AI intensives.

It wasn't glamorous. Open YouTube, check 20 channels I follow. Google a bunch of keywords. Skim articles. Copy stuff into a doc. Flag things for the newsletter. Four to five hours, give or take, every single week.

I didn't love it. But I figured it was part of the job.

That was before I built the two-agent setup that replaced all of it.

What the Setup Actually Looks Like

The workflow has two agents. Simple enough.

Agent 1 monitors 20 YouTube channels I care about. Whenever a new video drops, the agent summarizes it and sends that summary to a Slack channel. This runs automatically throughout the week without me touching anything.

Agent 2 is a researcher plus writer combo. Once a week it pulls all those Slack summaries together, identifies the most interesting stuff, and drafts the research section of my newsletter.

By the time Thursday rolls around, the research is basically done. I spend maybe 15-20 minutes reviewing it and making edits.

That's it. The five hours are gone.

I talked through the whole thing with Ilias, a structural engineer I coach who also runs an investment research newsletter on the side. He immediately asked: Can I just use Claude instead of Perplexity for the research part?

That question is important. Because the answer is why most people fail when they try to build something like this.

The Mistake: Forcing One AI to Do Everything

LLMs like Claude, ChatGPT, and Gemini are incredible at synthesizing information. They can take a pile of text and turn it into something clean, clear, and useful.

But finding information? Searching the internet in real time? That's not what they're built for.

Perplexity is a search tool. Google is a search tool. These are designed to surface what's actually on the internet right now.

LLMs are interpretation tools. They're built to take information you already have and do something smart with it.

When people try to use Claude or ChatGPT as a research tool, they're asking the wrong tool to do the job. The outputs feel off. The agent misses things. And then they blame the agent instead of the architecture.

I call this being Multi-Tool Native. The best AI users don't fall in love with one platform and try to force it to do everything. They treat different tools like specialists and route each task to whoever's best at it.

Here's roughly how I think about it:

• Perplexity: find things on the internet, real-time research
• ChatGPT or Claude: synthesize, draft, interpret, explain
• Lindy: automate recurring workflows, connect tools, run agents on a schedule
• Gemini: visual tasks, anything inside Google Workspace

Four tools. Four different jobs. They're not interchangeable.

The 80-20 Rule for Picking What to Automate First

I get asked a lot: Where do I start with AI automation?

My answer is always the same. Start with the thing you do most often.

Not the coolest use case. Not the most impressive demo. The thing that repeats.

Daily or weekly tasks are where the compounding kicks in. The newsletter research was a perfect candidate. It happened every single week without fail. Five hours every Friday. Over a year that's north of 200 hours.

I track how much time my agents save me every week. At my peak, I hit 83 hours in a single week. One week recently was 34 hours. The number one driver, every time I check, is email. My inbox agent handles the bulk of it.

But the research automation was the one that surprised me most. Because I didn't realize how much I was losing until it was gone.

How to Think About Building This for Yourself

You don't have to start with 20 YouTube channels and a two-agent pipeline.

Start with one repetitive information task. Something you gather, track, or summarize on a regular basis. Could be news in your industry. Could be competitor updates. Could be something specific to your client work.

Then think about two steps:

1. Find it. Use Perplexity or Google-based tools to surface the information automatically.
2. Interpret it. Use an LLM to summarize, extract key points, or draft something from it.

Most people skip step one or use the wrong tool for it. That's where the results fall apart.

Getting the tool routing right is 80% of the battle. Once that clicks, building the actual agent is the easy part.

The Takeaway

Five hours of weekly research. Gone.

Not because I found a magic AI tool that does everything. Because I stopped trying to make one tool do everything.

Perplexity finds. Claude interprets. Lindy automates. Each one doing the job it's actually built for.

If your AI setups keep disappointing you, ask: am I routing this to the right tool? Or am I forcing one tool to be all things?

Try it on your next repetitive research task. Build the find step and the interpret step separately. See what happens.

Want to see how to build something like this yourself? We cover agent architecture in our AI workshops. Details at asianefficiency.com.

The Bookshelf Problem

You know, when you walk into someone's home and see 200 books on their shelf? Your first thought is never “I need to read all of those this week.” You just go, wow, they've been around a while. They've had time.

That's the exact same thing as my AI collection.

I've been building since February 2025 — just over a year now. One agent at a time. Some worked great on the first try. Some took three rebuilds to get right. A few I deleted entirely because I thought I needed them, but did not.

The collection did not appear overnight. It accumulated.

Where I Actually Started

My first Lindy agent was an email drafter. Basic. Very basic. All it did was take emails from my inbox, pull context from previous conversations, and draft a reply for me to review.

Saved me maybe 20 minutes a day. That was the whole win.

That is it. I did not automate my calendar at the same time, or build a research pipeline, or set up weekly reports. I just had a thing that drafted email replies and I used it every morning.

Boring, right? But that 20 minutes compounded.

The One Tweak a Week Frame

I teach something called One Tweak a Week for AI adoption. The idea is dead simple: do one more thing with AI this week than you did last week. Not ten things. Not a full automation stack. One thing.

Week one: get an email drafter working. Week two: set up a meeting prep brief. Week three: add a basic weekly digest.

That pace feels slow. But it does not feel slow six months in when you look back at what you have built.

I run workshops in Austin every six weeks. I have watched people go from never using Lindy to running a 5-agent system in 90 days just by following this pacing.

The 55 Hours Moment

Lindy sends me a weekly email with my time savings report. At some point earlier this year, the report said I had saved 55 hours that week. I had to re-read it.

That is what most people miss when they look at someone's complex AI setup. They see the end state. They do not see the two hundred micro-experiments that got it there.

The AI Fluency Framework

When I teach workshops, I describe three stages of working with AI: AI Assisted, AI Workflows, and Building Agents.

Most people try to jump straight to stage three because that is what they see in demos. But you need stages one and two to make stage three work. The skills compound.

What to Do This Week

Pick one thing you do every week that is predictable and annoying. Something with a clear input and a clear output.

Build one agent that does that one thing. Use it for three weeks before you build anything else.

The bookshelf fills up faster than you think once you get going. But it fills up one book at a time.

One of the participants raised his hand about halfway through. He had built the whole thing. Connected his inbox. Set up the trigger. The agent was working, drafting replies. But every draft was different. Sometimes short and punchy. Sometimes three paragraphs long. No consistency.

He was frustrated. He asked me what was wrong with the prompt.

I looked at it. The prompt was fine. The instructions were clear. But there was one thing missing.

There was no example.

The Missing Piece Nobody Talks About

When you write an agent prompt, you are basically giving someone a job description. You tell them what you want, what to avoid, and how to handle edge cases.

But imagine hiring a new assistant and handing them a job description with no examples of good work. They would read every instruction carefully… and still produce wildly different results, because they are interpreting everything based on their own defaults.

AI agents do the same thing.

Without a reference point, the model fills in the gaps with its own judgment. Sometimes it guesses right. Sometimes it does not. And there is no way to know in advance which one you will get.

I call what fixes this the sample of success — a real example of the output you actually want, added to the bottom of the prompt.

What I Added (And What Happened)

After the participant showed me his prompt, I had him scroll to the bottom and add one email.

Not a hypothetical email. A real reply he had written that he was happy with. Short, direct, around 5 sentences. The kind of reply he would want his agent to write.

He ran the agent again.

The next five drafts? All came back in the 4-6 sentence range. Same tone. Same structure. Totally consistent.

The agent was not smarter. The model did not change. He just gave it something to aim for.

The difference between “write a professional email reply” and “write a professional email reply that looks like this…” is enormous. The second instruction does so much more work, even though it feels like a small addition.

This Applies to Every Kind of Agent

The sample of success is not just an email thing. It works for any agent where output quality or format matters.

Content agents: If you are using an agent to draft LinkedIn posts or newsletters, add one post or newsletter you have already written and liked. The agent will start mirroring that structure, that length, that tone.

Meeting summary agents: Add an example of a meeting summary in the exact format you want. Two sentences of context, then bullet points? Show it that. Three-paragraph narrative? Show it that.

Research agents: If you want a research brief in a specific format, paste in a good one. The agent will use it as a template without you having to describe the template in words.

The pattern is always the same: the clearer your example, the more consistent the output.

The Two-Part Prompt Formula

At my workshops, I teach something I call the OCE formula for agent prompts — Outcome, Context, Expectations. Most people get the outcome right. They state what they want. Some people add good context. But the Expectations piece… that is where the sample of success lives.

Expectations tell the agent: here is exactly what done looks like.

Without that piece, your prompt is incomplete. The agent knows the goal but not the standard.

And honestly? Most AI-generated content I have seen is weak for exactly this reason. Someone gave the model a goal without a standard. So the model filled in the standard itself, and it did not match what the person actually wanted.

The Shortcut When You Are Stuck on the Prompt Itself

Here is something that trips a lot of people up: they know they need a better prompt, but they do not know how to write one.

I have started using what I call meta prompting for this.

Instead of staring at a blank text box trying to craft the perfect prompt, I ask the AI to write it for me. I describe what I want the agent to do, paste in an example output if I have one, and say: “Write me an agent prompt that would produce something like this.”

Claude and ChatGPT are both good at this. What comes back usually includes a clear role, a goal, some context, a set of instructions, and — if you gave it an example — a sample of success already baked in.

A client of mine was struggling to write a prompt for a complex spreadsheet automation. Instead of coaching him through the whole prompt-writing process, I told him to take a screenshot of the spreadsheet, paste in what he was already trying to do, and ask Claude to write the agent prompt for him. He had a solid prompt in about 3 minutes. He had been stuck for an hour.

The AI can write its own instructions. You just have to know to ask.

Start Here

If you have an agent right now that is producing inconsistent output, here is what to try:

1. Pull up the prompt.
2. Find one real example of the output you would actually want to use.
3. Add it to the bottom of the prompt with a label like “Example output:” or “Sample of success:”
4. Run the agent again on the same input.

If the variance does not drop significantly, the prompt might need work. That is when meta prompting helps — paste the whole prompt into ChatGPT or Claude and ask it to improve it.

Inconsistency is not a model problem. It is an instruction problem. And it is usually fixable in about 10 minutes.

I cover this kind of thing in my one-day AI workshops in Austin — hands-on, no slides, real agents built by the end of the day. If that sounds useful, reach out.

Visit www.asianefficiency.com for more productivity tips and tactics.

Sign up for a $1/month trial period at shopify.com/tps.

Get the right life insurance for you and save more than 50% on term life insurance at SelectQuote.com/TPS.

Visit Upwork.com right now and post your job for free.

Cheat Sheet

• Top 3 Productivity Resources [2:25]
• The real reason you can't start that task (hint: it's not laziness) [4:33]
• How watching someone else succeed can unlock your own confidence [9:42]
• The “I'll just” trick that makes any task feel ridiculously easy to start [19:46]
• Why you'll never procrastinate on something that truly matters to you [21:20]
• The sprint mindset that made Brooks's personal velocity skyrocket [28:22]
• What your 30 open browser tabs are really costing you [33:54]

Links

• Shopify
• SelectQuote
• Upwork
• 25X Productivity Coaching
• The Procrastination Equation by Piers Steel
• The Code Book by Simon Singh
• Vitamix Wet Blade Replacement
• How to Change by Katy Milkman
• Claude
• The 12 Week Year by Brian P. Moran

If you enjoyed this episode, follow the podcast on Apple Podcasts, Spotify, Stitcher, Overcast, Pocket Casts or your favorite podcast player. It’s easy, you’ll get new episodes automatically, and it also helps the show. You can also leave a review!

We got to the part about conditional routing. “If the meeting is external, do X. If the attendee is on your VIP list, do Y. If it's a recurring sync, skip the prep email.”

One of the participants, Mark, kind of laughed and said he was old enough to remember doing actual flowcharts back in the day. He said it almost apologetically, like he was dating himself.

I stopped the session. “Mark,” I said. “That makes you more qualified for this than you think.”

What AI Agents Are Actually Made Of

Here's the thing about building AI agents that nobody explains clearly enough.

It's two things: logic and prompting.

Prompting is the part everyone talks about. How to write a good instruction. What context to give the model. How to phrase the output format. That stuff is learnable and you can get decent at it in a weekend.

The logic part is different. That's the “if this, then that” decision structure underneath the agent. It's what determines whether an agent is actually useful or just a glorified auto-reply.

Most people who try to build AI agents get stuck on the logic layer. Not because they aren't smart. Because they've never had to think through workflows in that structured, conditional way before.

But ops people? Project managers? Consultants who've built SOPs and process maps?

They've been doing that exact thinking for years.

Why Your Old Skills Transfer Directly

I've been thinking a lot about skill stacking lately. The idea that combining two skills creates something rare that most people can't replicate.

I've seen this at Asian Efficiency. Brooks, our operations guy, also learned digital marketing. Neither skill alone made him uniquely valuable. But an ops person who deeply understands digital marketing campaigns? That's genuinely hard to find.

Same principle is playing out right now with AI agents.

If you have experience writing SOPs and process documentation, practice thinking through “what happens in each case” logic, and a habit of mapping workflows before automating them… you have the primary ingredient for agent building.

The prompting layer is learnable. The logic layer takes much longer to internalize if you're starting from scratch. Most people are starting from scratch.

You're not.

What This Looks Like in Practice

Let me make this concrete.

A calendar agent that handles inbound meeting requests needs to answer questions like:

• Is this person already on my contact list?
• Is this a request for a specific time or an open-ended “let's meet”?
• Does this conflict with anything already blocked?
• What's the appropriate response if no slots are available this week?

That's a decision tree. It has branches. Some branches lead to different actions.

If you've ever built a flowchart, you know exactly how to think through that. You've probably built more complex logic than that for an actual operations process.

The only difference is that instead of drawing boxes and arrows, you're now writing it in plain English inside an agent prompt. That's not a dramatic skill leap. That's a translation.

Where People Get This Wrong

There's a common assumption floating around that AI is mostly for developers and tech people. That if you don't have a technical background, you're behind.

I push back on that hard.

The best agent builders I've worked with aren't necessarily technical. They're people who think clearly about process. They know how to decompose a workflow into steps. They can articulate edge cases. They understand what “done” looks like for a task.

That's not a developer skill. That's an ops skill. And it transfers almost directly.

What technical people sometimes lack is the habit of asking “what happens in each case?” before building anything. That single habit is worth more in agent design than any programming language.

The Skill Nobody's Talking About

Most AI content right now is focused on prompting tips, tool comparisons, and productivity hacks. That's fine. But it misses something important.

The people who are going to be exceptional at building agents are going to have strong process fundamentals underneath. They're going to look at a workflow and immediately see the decision points. The exceptions. The places where an agent would need to ask a clarifying question vs. just take action.

That's process mapping thinking.

And if you've spent years as an ops person, a project manager, a consultant, or anyone who's had to document how work actually flows through an organization… you already have that.

You just didn't know it was relevant yet.

It is.

Want to build your first AI agent without needing a technical background? Join an upcoming workshop to see this in action.

And I decided to do something unrehearsed.

I pulled up Lindy on the big screen, hit the voice button, and said: “Hey, check the sentiment around Red Ash — the restaurant here in Austin. If it's positive, find a time on my calendar and add it as a walk-in.”

One sentence. No other instructions.

Then I turned to the group and said, “Watch this.”

What Happened Next

Lindy went to Perplexity first. It searched for reviews and sentiment about Red Ash. Came back: overwhelmingly positive.

So it kept going.

It went to Google and looked up Red Ash's opening hours.

Then it opened my Google Calendar and found a free slot that matched those hours.

Then it added a calendar event. Automatically.

The room went quiet. Not the polite quiet of people being respectful — the actual quiet of people whose brains just caught up to what they witnessed.

I said “I'm going to delete this because I actually don't feel like going right now,” and a few people laughed. But then someone asked: “How many tools did that use?”

Four. It used four separate tools, chained together, from a single sentence. No additional prompts. No clicks.

Why This Isn't Just a Party Trick

Here's what I want to be clear about: that demo wasn't meant to impress. It was meant to show the difference between using AI as a tool and using AI as an agent.

Most people are still at what I call the AI Assisted stage. They open ChatGPT, ask a question, get an answer, copy it somewhere. That's fine. That's genuinely useful. But it's basically a faster, smarter Google search.

When you move to the agent level, something different happens. The agent doesn't just answer — it reasons across multiple sources, makes intermediate decisions, and takes action inside real systems. No babysitting required.

The Red Ash example hit four tools in sequence:

1. Perplexity (real-time research for sentiment)
2. Google (look up hours — actual business data)
3. Google Calendar (find availability)
4. Google Calendar again (create the event)

Each step required a decision. “Is the sentiment positive enough to continue?” “Do the hours overlap with Thanh's free time?” “Which slot makes sense?”

A chatbot can't do that. A chatbot waits for you to tell it what to do next.

The Cost Question

At the end of that session, someone asked how much all of that cost.

I pulled up the Lindy usage dashboard and checked. The entire session — multiple scheduling tasks, calendar rescheduling, the restaurant lookup, all of it — ran about 15 cents.

Think about that against the alternative: emailing an assistant, waiting 30 minutes to an hour, getting the calendar invite, reviewing it, confirming back. That's a real-world workflow for a lot of people.

15 cents. Done in seconds.

I'm not saying this to be dramatic about it. I say it because people underestimate the economic argument for agents. Once you see the usage cost on something like this, it changes how you think about what's worth automating.

The Mental Model Shift

Here's what I find most useful for people just getting into this: stop thinking about AI as a question-answering machine.

Start thinking about it as a digital teammate that can use tools.

You give it an outcome. It figures out the steps. It uses the right tool for each step. It reports back when it's done.

That's the agent model. And when you internalize that model, you start seeing automation opportunities everywhere. Not just “I should ask ChatGPT about this” — but “I could build a workflow that handles this every time it comes up.”

The multi-tool chaining is what makes agents genuinely different. The ability to use Perplexity for research, Google for facts, a calendar for scheduling, an email system for communication — and chain all of those together inside a single instruction.

That's not a better chatbot. That's a different category.

Where to Start

If you're still mostly using AI for one-off questions, that's a great starting point. But the next level is learning to think in workflows.

Pick one task you do every week that involves at least two steps. Map out what a human would do to complete it. Then ask whether an agent could handle those steps.

If the steps involve information lookup, any kind of scheduling, or sending a message to someone — there's a good chance an agent can take it off your plate.

The Red Ash demo was unrehearsed because I wanted people to see how natural it is once you've built that muscle. One sentence. Four tools. Done.

That's the goal.

Thanh runs hands-on AI workshops in Austin for entrepreneurs and business owners. If you want to see what agentic workflows look like for your specific situation, check out the next available workshop.

I did about five minutes of explaining how our super agent worked. They listened. Nodded. Asked a couple of questions.

Then I stopped talking and just ran it. Pulled up the agent, typed a quick prompt, and it generated a full email with a linked Google Doc in about 30 seconds.

The son stayed quiet. But the dad said something I've heard variations of dozens of times now.

“I'm currently thinking of 30 or 40 different things I could do to implement this.”

Thirty to forty things. Immediately. From a single five-minute demo.

That's not a great pitch deck or a clever case study doing that. That's the thing itself, running in real time, using real context.

The Difference Between Explaining and Showing

Here's what I've noticed after doing AI workshops across Austin for most of the past year.

When I explain how an AI agent works, people follow along. They nod. They say “that's interesting.” And then we move on and they forget most of it by the next morning.

When I show an AI agent working on their actual business… something different happens.

I did a session where I walked someone through a SWOT analysis using ChatGPT. Not their imaginary business. Their real business, with their real data. Within about 10 minutes, hands were up across the room. People wanted to know what else it could do. They weren't asking if AI was useful. They were already past that question.

The demo had done something explanation couldn't: it gave them permission to imagine.

Why Live Demos Work the Way They Do

There's a gap between understanding that AI exists and understanding that AI applies to your specific situation. Explanation closes the first gap. Demos close the second.

When someone watches a live demo using their own industry, their own terminology, their own problems… the brain starts making connections automatically. Jacob's dad wasn't waiting for me to tell him what to do with the tool. He was already 30 steps ahead in his own head.

This is the whole design principle behind the workshops I run. I teach a three-level progression… AI Assisted, AI Workflows, and Building Agents. But the way I get people to believe that progression is real isn't by showing them a slide about it. It's by running something live and letting them see what level three looks like in practice.

Once they see it, they stop asking “is this worth it?” They start asking “how do I get there?”

A Finance Team That Went From Skeptical to All-In

I worked with a finance and accounting team at a real estate company earlier this year. Standard situation: they were curious about AI but not convinced.

I did a demo two weeks before the tools we were discussing even officially launched. Showed them what was coming and how it would fit their workflows.

When the tools actually came out? The team went crazy. Not because of a great rollout or a polished training deck. Because they'd already seen it in action and spent two weeks thinking about what they'd do with it.

The demo had done the convincing weeks earlier. Everything after that was just logistics.

How to Run a Good Demo

Use their actual stuff. If you can load their data, their emails, their business name… do it. Generic examples feel like commercials. Real examples feel like magic.

Keep it short. Five to ten minutes is enough. You're not trying to show everything. You're trying to generate the “30-40 things” moment. Let them fill in the rest.

Run it live. Pre-recorded demos don't have the same effect. Something about watching it happen in real time makes it feel more real.

Let silence happen. After the demo, don't immediately start talking. Give people a second to process. That's when the ideas come.

The Pitch You Don't Have to Make

When a demo works, you don't have to pitch anything.

Jacob's dad wasn't asking me to convince him AI was worth it. He was already sold. He was figuring out the implementation in his head before I even finished the demo.

That's the goal. Not to make a compelling argument. To show something real and let people's imagination do the rest.

If you're trying to get your team, your clients, or your leadership on board with AI… stop rehearsing your pitch deck. Fire up a demo. Use real data. Run it live.

The ideas will come. You just have to get out of the way.

It was just me and a buddy catching up on some project stuff.

Three bots. Two people. Completely absurd.

But something clicked while I was laughing about it: I realized the transcript was the least interesting part. What mattered was what happened after the transcript. And for most people, what happens after is… nothing.

They get the summary email. Maybe they skim the action items. Then the whole conversation — every insight, every story, every aha moment — disappears into a folder nobody opens again.

That's the gap. And it's a big one.

The Conversation You're Already Having

A few weeks ago, I had a call with Monica, who runs a nonprofit here in Austin called CC4C. She was exploring AI tools — how they could help her organization run more efficiently.

At one point she mentioned how often her staff shares powerful donor impact stories in internal team meetings. Not formal presentations. Just: someone on a call describing how a specific donation had changed a family's situation. How a program finally worked. How a kid got access to something they never would have had otherwise.

I asked her: does any of that ever make it to LinkedIn? To a newsletter? To a grant application?

She paused. “No, not really.”

That's when I said something that kind of stopped the conversation: those stories are already content. The raw material exists. Someone said it out loud on a recorded call. All you're missing is a system to catch it.

What Most People Do With Transcripts

If you're using a meeting notetaker — Lindy, Otter, Fireflies, Granola, whatever — you're already capturing the raw text. The question is what you do with it.

Most people use transcripts for:

• Action items
• Decision logs
• Meeting summaries to share with people who weren't there

That's fine. Those are real uses. But it's the lowest-value thing you can extract from a transcript.

The higher-value extraction is stories. Insights. Moments. Things someone said that were unexpectedly good.

A client describing what changed in their business after using your product. A question from a workshop attendee that reveals a gap in how the market thinks about your category. A funny thing that happened that actually illustrates a serious point.

These are the pieces that turn into content people actually want to read.

The Transcript First Approach

I've been running what I call a “transcript first” workflow for about a year now. Everything I record — coaching calls, client demos, workshops, community sessions — gets processed through the same system.

The basic idea: every conversation generates assets. Not just notes. Actual reusable material.

The workflow has a few steps:

1. Capture everything. This is the non-negotiable. If you're not recording and transcribing, you have nothing to work with. Get a meeting bot set up and running on every call where there's a consent-appropriate reason to record.

2. Extract stories, not just facts. Most notetaker prompts are configured to pull action items and decisions. Reconfigure them (or add a second pass) to also pull: interesting moments, unusual insights, stories someone shared, results or before/after comparisons.

3. Route stories to a content doc. I keep a running Google Documents where my content pipeline drops extracted story material. It's not polished. It's raw. But it's searchable and reusable.

4. Let AI write the first draft in your voice. Once the raw story is in the doc, you can hand it to an LLM with your voice profile and ask it to turn it into a LinkedIn post, a newsletter paragraph, or a section of a longer article.

The whole thing is automated. A meeting ends. The transcript gets processed. Story material gets flagged. It drops into the content doc. I review and edit. Done.

Why This Works Better Than “Writing Content”

The hard part of content creation isn't writing. It's finding something worth saying.

Most people sit down to “write a LinkedIn post” and stare at a blank screen because they're starting from zero. They're trying to generate an idea from nothing.

The transcript approach flips this. You're not generating ideas. You're finding them in conversations you already had.

And conversations are almost always more interesting than ideas you invent alone. Real stories beat thought leadership. Specific examples beat general advice. What someone actually said beats what you think sounds smart.

When I told Monica that her team's internal stories were already content, she immediately understood. The bottleneck wasn't creativity. It was infrastructure. She had everything she needed. She just wasn't routing it anywhere.

A Practical Setup You Can Build This Week

If you want to try this, here's the simplest version:

Start with one week of calls. Pick your AI notetaker and add a secondary prompt: “In addition to action items, list any stories, examples, or surprising moments from this conversation. Include who said it and a one-sentence context.”

At the end of the week, collect what came out. You'll probably find five to ten moments across your calls that could become content.

Take the best one. Hand it to ChatGPT or Claude with a simple prompt: “Write a LinkedIn post in my voice based on this story. Keep it casual and specific. No em dashes. Short sentences.”

Edit it. Post it. That's the proof of concept.

Once you've done it once, you'll start seeing every meeting differently. Not just as a task to get through. As a potential asset.

The Nonprofit Story Ended Well

At the end of my call with Monica, she started brainstorming immediately. She mentioned three or four specific stories her team had shared in recent meetings that she'd completely forgotten about until we started talking.

One of them — about a teenager who'd gotten into college partly because of a program CC4C ran — would make an incredible newsletter story. She already had the transcript. She already had the permission to use it.

She just needed the system to make it happen.

That's all most people need. Not more content ideas. Just a better way to catch the content they're already creating.

Want to build a system like this? Check out the Productivity Academy — we cover AI workflows for content creation, meeting intelligence, and more.

She’s good at what she does. Custom menus, dietary restrictions tracked, clients with strong preferences about spice levels and proteins, and what they ate last week. Real relationship-based work.

And every single week, she was spending four hours writing menus by hand.

I asked her what that process looked like. She’d open her client list, pull up each person’s notes, think through what she’d made for them recently, factor in what was seasonal or on sale, and then type out a custom menu for each client from scratch.

Every week. Same process. Four hours gone.

When she said that, I didn’t immediately say, “Here’s the AI tool for that.” I asked more questions first. What format do the menus go out in? Where do you keep the client preferences? How much does the menu vary week to week?

That diagnostic step matters more than most people think. It’s the difference between building something that works and building something that looks good in a demo but fails on Tuesday.

The First Attempt (And Why It Broke)

Michelle had already tried to fix this herself before we talked.

She built a ChatGPT prompt that was supposed to do everything: read the client preferences, remember past menus, generate new ones, and export them to CSV all at once.

It kept failing. Forgetting which client was which. Messing up the CSV. Generating the wrong proteins for someone with dietary restrictions.

She figured she’d done something wrong with the prompt. But the prompt was fine. The architecture was the problem.

One agent trying to do four separate jobs at once is like asking someone to be the chef, waiter, cashier, and dishwasher simultaneously. They’ll do all of it badly. There’s a design principle I use with every client now: one agent, one job.

When you try to cram too much into a single AI prompt, you’re overloading the context window, creating conflicting instructions, and making it nearly impossible for the model to know which task should take priority when they conflict. The result is exactly what Michelle experienced — inconsistent, unreliable output that you can’t trust.

What We Actually Built

We spent about 90 minutes mapping the workflow together before touching any AI tool.

First, we figured out the data. She already had a spreadsheet with client preference tabs — one tab per client, with notes on their allergies, favorite proteins, what they’d eaten recently, and any restrictions. That spreadsheet was the source of truth we needed.

Then we designed backwards from the output. What does a finished menu actually look like? How long is it? What format does it need to be in? Once we knew exactly what “done” looked like, building toward it got a lot simpler.

The workflow ended up with two focused steps:

1. Read the client preference tab and pull the relevant data
2. Generate a personalized weekly menu based on those preferences

That’s it. No CSV export in the same step. No memory of past menus baked into the same prompt. Just two clean, focused tasks.

Michelle reviews the output, makes any adjustments that feel off (she still knows her clients better than any AI does), and sends them out.

The whole thing now takes about 30 minutes instead of four hours.

The Dual-Brand Problem

Michelle also runs two different lines of business under her name — a luxury catering arm and the personal chef meal service. They have completely different voices. The catering side is formal, polished, upscale. The personal chef side is warmer, more casual, like a friend who happens to cook for you.

She’d been writing all her client communications by hand to manage that tonal difference. Which made sense — she just didn’t have another option.

So we added a second piece to the workflow: she used ChatGPT to analyze her existing content for each line of business and write a brand voice prompt that captured the tone. Then we imported those prompts into her AI agent’s system settings.

Now when the agent generates a menu or a client update for the catering side, it automatically writes in the right voice. Same for the personal chef side. Two brands. Two distinct voices. Zero extra effort on her end.

This “use AI to write the prompt for another AI” approach sounds a bit meta, but it works really well for service businesses with multiple client segments or brand voices.

What Made This Actually Work

A few things made this project go smoothly that I want to call out because they’re not obvious.

She had clean data. The client preference spreadsheet was organized and up to date. If that hadn’t existed, we would have spent the first hour just figuring out where her client information lived. Data centralization isn’t glamorous, but it’s what makes automation possible.

We didn’t start with the tool. We started with the workflow. What’s the desired output? What’s the trigger? What are the steps a human would take? I call this designing backwards — start at the end state and work backward. It keeps you from building a solution in search of a problem.

She stayed in the loop. Michelle still reviews every menu before it goes out. The AI isn’t making final calls — it’s doing the drafting work. That human review step means clients still get her judgment, and she catches anything the AI gets wrong before it matters.

That last point is worth sitting with. The goal wasn’t to remove her from the process. It was to remove the part of the process that was just mechanical repetition, so she could focus on the part that actually requires her expertise.

Your Version of This

Almost every service business has a version of Michelle’s four-hour menu writing task. Some repeating process that requires your knowledge, but not your active presence every single time.

It might be client intake forms. Proposal generation. Weekly check-in emails. Social media captions. Progress reports.

The pattern is almost always the same: you know exactly what needs to go in, you’ve done it dozens of times, and yet you’re still sitting there typing it from scratch every week.

The fix usually isn’t complicated. It’s mostly about designing the workflow right before touching any AI tool.

If you’re curious what this might look like for your business, I run AI workshops where we actually build these systems together — not just talk about them. Check out my upcoming workshops at asianefficiency.com and see if there’s one that fits.

Or just try mapping your own version of “the four-hour task” this week. Write down the steps a human would take. Figure out where the data lives. Then ask yourself: which of these steps actually needs me?

That question alone might get you most of the way there.

We set up follow-up emails. Timed outreach after every appointment. A sequence that triggered automatically when someone left the chair. Looked great on paper.

Then we started testing it.

And that’s when I realized we’d been thinking about it wrong.

The Part We Almost Missed

The stylist was still the most important variable.

Not the email. Not the timing. Not the copy.

The person who cut the client’s hair.

If the stylist built a real connection during the appointment… the review almost always came. If the appointment was transactional and the client walked out feeling like a number… the automated follow-up hit the trash folder.

We had to go back and rework the whole process. Instead of treating AI as the driver, we treated it as support. The stylist does the relationship work. The AI handles everything that happens after.

That’s the actual split.

What I Keep Seeing in Service Businesses

I gave a talk in New Orleans last year to a room full of salon owners. Sirius Business conference. Three standing-room-only sessions.

I showed them things they’d never seen before. AI responding to text messages. Rebooking clients automatically when a regular hadn’t been in 90 days. Taking 500 unpublished product photos and sorting, cleaning, and resizing them in an afternoon.

The reactions were what I’d come to expect. Jaws dropped. “I didn’t know this was possible.”

But the question that comes after is always the same.

One owner pulled me aside after the session: “So what happens to my front desk person?”

I’ve gotten that question dozens of times now. From salon owners, from attorneys, from real estate brokers, from nonprofit directors.

And the answer is always the same: if you deploy this right, your front desk person gets promoted.

The Actual Role AI Plays in a Service Business

Here’s what I mean.

In a salon, the front desk person typically spends her day on three things:

1. Answering phones and booking appointments
2. Sending reminder texts and confirmation emails
3. Chasing no-shows and rebooking cancellations

That’s all automatable. Lindy can handle inbound texts, trigger reminders at the right intervals, and even draft rebooking messages when someone goes quiet.

So what does the front desk person do when that work is gone?

She does the thing AI genuinely can’t do. She learns every regular client’s name. She notices when someone walks in looking stressed and adjusts the energy. She makes a first-time client feel like they’ve been coming there for years.

That’s not a job at risk. That’s a job that gets better.

This is what I call “human in the loop” thinking. AI handles the repeatable work. The human stays in the loop for anything that requires judgment, trust, or relationship.

Why Most AI Deployments in Service Businesses Fail

The failures I’ve seen come from the same mistake: treating AI like a replacement instead of support.

A salon that auto-sends review requests without training stylists to ask in person… gets a fraction of the reviews they should.

A medical office that automates all their patient follow-up without ever explaining it to their staff… gets pushback and confusion on day one.

A real estate firm that builds an AI lead enrichment system but never changes how their brokers use their prep time… doesn’t see the ROI.

The technology works. The human integration doesn’t.

I built a briefing system for a VC named Evan. Before every call, he gets a prep doc. Context on the company, relevant news, old email threads, a suggested talking point or two. What used to take a human assistant 20 hours a week to put together now runs automatically overnight.

But Evan still does every call himself. He still makes every investment decision. The AI made him more prepared. It didn’t make him less necessary.

How to Get the Split Right

If you’re in a service business thinking about AI, here’s how I’d approach it.

First, map what your team actually does in a week. Break it into two buckets: things that require human judgment and relationship, and things that are just process.

The process stuff is where AI wins. Scheduling, follow-up, document prep, data entry, research, reminders. These are things that are currently eating your team’s best hours and delivering nothing extra because of it.

The human stuff is where your team should be spending more time, not less. Client relationships. Problem-solving. Handling anything sensitive or unusual. The moments that make or break whether someone comes back.

Second, deploy AI in the process bucket first. Get a win. Let your team see that it works and that it doesn’t eliminate them… it frees them.

Third, watch what your team does with the extra time. That’s where you’ll find the real ROI. Not in the hours saved. In what those hours get redeployed toward.

The Stylist Still Has the Most Important Job

When I went back and retooled the salon review system, we added one thing the AI couldn’t do.

We trained the stylists to ask.

Not a script. Just a natural mention near the end of the appointment: “If you had a good experience today, a review would really mean a lot to us.”

The automated follow-up email still goes out. But now it lands with someone who’s already been asked, who already had a great experience, who already feels like the stylist valued them.

The conversion rate on reviews jumped.

The AI didn’t do that. The stylist did. The AI just made sure the follow-through happened every single time, without anyone having to remember.

That’s the combination that works.

If you’re curious how to find those same opportunities in your business, I run workshops on exactly this. One day, hands-on, specific to your industry. Reply here or check the site for upcoming dates.

Thanh Pham is an AI consultant and workshop instructor based in Austin, TX. He’s helped businesses across salon, real estate, healthcare, and professional services deploy practical AI systems.

Michelle was spending four hours every single week writing menus by hand. She had ten clients, each with their own list of proteins, carbs, allergies, spice preferences, and weekly modifications. Some wanted rice, others wanted melon salad. Some were paleo. Some were dairy-free, but only for certain things.

She'd already built a custom GPT to handle it. She called it “Menu Maestro.”

And it was a disaster.

Every session, it forgot someone's profile. Every session, the CSV export failed. She kept having to start over, re-explaining the same preferences, getting halfway decent results before the whole thing broke down.

“He's missing clients,” she said, pulling up the screen. “Again.”

She thought the tool was broken.

It wasn't. The design was.

The Most Common AI Mistake I See

When most people build their first AI workflow, they do what feels logical: they give it the full picture and expect it to handle everything.

One GPT. One prompt. Ten tasks.

The problem is that language models aren't multi-taskers. They're not like a human assistant who can juggle context across fifteen different things and still remember that Janet is lactose intolerant.

A GPT does one job with high consistency. It does five jobs with mediocre results across all of them.

This is what I call the “one agent, one job” rule. It's the single most important design principle for building AI workflows that actually work.

Michelle's Menu Maestro was trying to:

1. Remember ten client profiles
2. Generate a master menu
3. Customize each dish per client
4. Flag exceptions and modifications
5. Export everything to a Google Sheet

That's five separate jobs. And every job it had to do increased the chances of it dropping the ball on the others.

What Good Design Looks Like

Once I understood what was happening, the fix was straightforward.

Instead of one GPT doing everything, we broke it apart:

Step one is understanding the client. A dedicated context file per client — stored as a spreadsheet tab — that the GPT reads at the start. Not a “remember this” prompt buried in a conversation. A structured file it can reference reliably.

Step two is generating the base menu. One prompt, one output: here's the master set of dishes for this week.

Step three is the customization. For each client, take the base and apply their preferences. One client, one output. No trying to do all ten at once.

That's it. Three focused steps. Chain them in sequence.

“I would cry with happiness,” Michelle said when we mapped it out.

We estimated her four-hour weekly process could drop to about thirty minutes once the workflow was running properly.

Why You Should Start With ChatGPT, Not Lindy

Here's a counterintuitive thing I tell people who want to build AI automations: don't start with the automation tool.

Start with ChatGPT.

When Michelle mentioned the session was supposed to be a Lindy workshop, she asked — shouldn't we be building this in Lindy?

Not yet.

Lindy is excellent for external integrations: connecting to email, calendar, Slack, Google Sheets, triggering workflows automatically. But it's not the fastest tool for prototyping the actual logic of what you're building.

ChatGPT lets you iterate the design in minutes. You paste a prompt, see what it produces, adjust, try again. Once the logic works reliably in chat, porting it to Lindy is easy.

This is what I call “start small, iterate.” The smallest useful version of any workflow isn't the automated version. It's the working version.

Automation before validation is how you end up building a complicated system that reliably produces the wrong output.

Prove the logic first. Then automate.

A Framework You Can Use Right Now

Before you build your next AI workflow, answer these three questions for each step:

What is the one output this step needs to produce?

Not “manage the whole process.” One output. “Generate a list of ten menu ideas based on these ingredients and these client restrictions.” One thing.

What is the minimum input it needs to do that job?

Client preferences, weekly ingredients, and nothing else. Don't feed it the entire history of the business. The more you add, the less reliable it gets.

Can this step succeed without knowing what comes next?

If yes, it's a good candidate for a standalone step. If it needs to “keep track” of what it's going to do five steps later, it's doing too much.

Three questions. If each step can answer them clearly, your workflow will be more reliable, easier to debug, and faster to build.

The Lesson That Applies Beyond AI

What I keep seeing in workshops and client sessions is that most people approach AI with a project management mindset. Define the big outcome, build toward it, see if it works.

But AI workflows are more like cooking.

You don't throw everything in the pot at once and hope it turns out. You have a sequence. You build flavor in stages. One step informs the next.

Michelle is a personal chef. She would never give someone a raw list of ingredients and say “figure it out.” She designs the week's menu first, then customizes per client, then writes the shopping list.

The AI workflow for her business just needed to mirror how she already worked.

That's the principle: design your AI workflow the way you'd design the process itself. Break it into steps. Give each step one job. Then chain them together.

The tool isn't the problem. The design almost always is.

Want to learn how to build AI workflows that actually work for your business? Thanh runs hands-on AI workshops in Austin — one day, 9am to 5pm, personalized by industry.

We were exploring a potential partnership. She wanted to bring AI tools to her audience. I wanted to understand who her people actually were before I started pitching anything.

She described them in a way I keep thinking about:

“They are all extremely not tech-savvy. They would want someone to build it. They would hire it out. They are very good at revenue. Very good at sales. Very good at serving their clients. But they are like pen and paper. Some of them don't even know how to add a Google Calendar.”

I've heard versions of this from a lot of people I've talked to over the past year. And I think it reveals something important about where the real AI opportunity is right now.

The Myth of the “AI-Ready” Client

When most people talk about selling AI services, they picture someone like themselves. Someone who's already tinkering with ChatGPT, reads the newsletters, and watches the demos. Someone who just needs help going deeper.

But that's a pretty small market.

The much larger market is people running real businesses who have zero interest in learning a new skill set. They're already stretched. They're good at what they do. A salon owner with three locations isn't lying awake at night thinking about agents and context windows. She's thinking about her stylists, her retention numbers, and whether the new Aveda line is worth the shelf space.

That's not ignorance. That's good prioritization.

And here's the thing: those clients often get better results from AI than the tinkerers do. Not because they're smarter, but because they're more focused. They hire someone to solve a specific problem, that problem gets solved, and then they go back to what they're actually good at.

The Education Phase Nobody Skips

Daniela had been working with her audience for three to four months before she even started talking about selling AI services. She called it “seeding.”

Just teaching people what the language means. What does “agent” mean? What does “automation” actually look like? What can it do?

This sounds slow. But it's the right order.

When I teach AI workshops, I see a version of this every single time. People come in skeptical. They've heard the hype. They don't know what applies to them. The first hour isn't really about AI at all… it's about recalibrating expectations.

The moment it clicks is usually during a live demo. I'll pull up a real workflow and show someone their exact problem being solved in real time. Email drafted in 12 seconds. Meeting prep done automatically. Research that would take an afternoon done in minutes.

The room changes. I watched someone in one of my February workshops go from folded arms to rapidly taking notes in about 90 seconds. That's not a product decision. That's a “don't know what you don't know” moment resolving in real time.

But that moment only works if someone's been warmed up to the possibility first. You can't skip to the demo without the foundation.

Two Markets, Two Very Different Services

There's a real split in who buys AI services right now.

One group wants to learn. They want to understand how it works. They'll watch your course, attend your workshop, try things themselves. They're excited by the process. This is the DIY market.

The other group wants it done. They don't want to learn prompting. They don't want to think about which tool is best for which task. They want someone to build the thing, train them on the one button they need to push, and then be available when something breaks. This is the done-for-you market.

Neither is wrong. But they're different businesses.

The DIY market is where workshops, courses, and content marketing work well. You educate, demonstrate, and let people self-select into your programs.

The done-for-you market is won with trust. Specifically, with trust built over time before you ever ask for money. Daniela's three-month seeding window wasn't a marketing delay… it was the marketing.

What High-Touch Actually Means

One thing I do with every client: I give them my personal cell.

When I told Daniela this, she paused. “I don't know if you should do that with our people.”

I understand the concern. But this is actually a feature, not a risk.

When you're selling AI services to non-technical clients, the biggest fear isn't the price. It's the feeling of being stranded. What if it breaks and I can't fix it? What if I don't understand it six months from now?

Having direct access to the person who built it removes that fear. And in my experience, people almost never abuse it. They text occasionally with a quick question, feel reassured, and keep using the system. That access is part of what they're paying for.

This also creates referrals. Non-technical clients talk to other non-technical clients. Word spreads that you're actually available. That's a stronger acquisition channel than any ad campaign.

The Right Framework for This Work

I use what I call the AI Fluency Levels to think about where any client is starting from. Level 1 is AI-Assisted: using AI tools directly, like ChatGPT for writing or Perplexity for research. Level 2 is AI Workflows: stringing tools together to automate repeating tasks. Level 3 is Building Agents: creating autonomous systems that run without you.

Most non-technical clients don't need to reach Level 3 themselves. They need someone who's already there to build Level 3 for them. And they need enough Level 1 understanding to feel confident using what gets built.

The education gap Daniela described is real. But closing it doesn't require turning a salon owner into an AI builder. It just requires showing her enough to trust the outcome.

That's a very different job. And honestly… a more interesting one.

Thanh Pham runs AI workshops for business owners and consults on AI implementation. If you want to explore what's possible for your business, learn about his workshops here.

Visit www.asianefficiency.com for more productivity tips and tactics.

Go to HomeServe.com to find the plan that's right for you. Not available everywhere. Most plans range between $4.99 to $11.99 a month your first year. Terms apply on covered repairs.

Links

• HomeServe
• 25X Productivity Coaching
• ChatGPT
• Claude
• Anthropic
• Gemini
• Claude Cowork

If you enjoyed this episode, follow the podcast on Apple Podcasts, Spotify, Stitcher, Overcast, Pocket Casts or your favorite podcast player. It’s easy, you’ll get new episodes automatically, and it also helps the show. You can also leave a review!

She told me something I keep thinking about: “In the spa world, everybody is essentially stuck in the nineties.”

Spa software hasn't changed in decades. Most spa owners don't know how to add a Google Calendar event. And Daniela, who's been studying AI for two years and is considered the AI-forward voice in her space, says she's still a beginner compared to what people in tech circles are doing.

That gap right there? That's the opportunity.

The Industries Furthest Behind Are the Most Interesting

When people ask me about AI adoption, they usually assume the interesting things are happening in tech companies, startups, and Silicon Valley offices.

But I've been teaching AI workshops to salon owners, spa owners, real estate agents, and contractors. And every single time I walk into one of those rooms, I see the same thing.

Not resistance. Not skepticism. Just… they had no idea this was possible.

At a salon conference a while back, all three of my sessions were standing room only. These were business owners who didn't know AI could respond to their text messages. Didn't know it could handle phone calls. Didn't know a 30-minute intake conversation could generate 30 days of social content automatically.

The moment they see it, they want it immediately.

This is what I mean by the first-mover advantage in non-tech industries. In these spaces, nobody's doing this yet. The person who shows up first with real, working AI solutions becomes the go-to expert almost overnight.

What Being Behind on AI Actually Means

Here's a useful reframe. When I say these industries are behind, I'm not talking about people who are lazy or resistant to change.

Spa owners are incredible at their craft. They're excellent at revenue and sales and serving clients. They just happen to run businesses where the technology layer has been mostly ignored for twenty years.

The same is true for contractors, lawyers, physical therapists, and most of healthcare. These are people who went deep on their actual profession. The technology side just never demanded their attention until now.

And because of that, almost nobody in their industry is ahead of them on AI. The playing field is flat.

The approach I call “one tweak a week” applies perfectly here. You don't need to leap from zero to full automation overnight. You pick one thing AI can help with this week, implement it, see the result. Then you add the next tweak next week. Let curiosity compound.

Where to Start: Revenue First

Here's the thing I keep telling business owners in these industries. Start closest to revenue.

If you can show a spa owner that AI will generate more five-star reviews, she'll be interested. If you can show her that AI will help her post on social media every day without hiring someone, she'll be very interested.

But if you lead with “AI can optimize your scheduling” or “AI can cut your operational overhead”… you've lost her.

I've seen this pattern with multiple clients. The ones who start with revenue-generating applications get hooked immediately. And then, once they've seen it working, they come back asking about everything else. The operations, the payroll, the scheduling. But you have to earn that trust first with something that visibly moves the needle.

This is why I built a content strategy agent for a jewelry business recently. Here's how it works:

1. A 30-minute intake call where we capture the brand voice and content strategy
2. The agent ingests all of that and generates a 30-day content calendar in Airtable
3. Every day, it posts automatically. The owner reviews and approves once a month to make sure everything still looks on brand

Total cost: about $6,000 to build. Then $50-100 a month to keep running.

For context, a social media manager typically costs $3,000-5,000 per month. You can do the math.

The Part AI Cannot Replace

One salon owner I worked with wanted more five-star reviews. Makes sense, they're critical for local businesses.

We built an automated follow-up system. But something interesting happened when we dug into it. The biggest driver of reviews wasn't the follow-up emails. It was the hairstylist's relationship with the client during the appointment.

AI could automate the ask. But it couldn't automate the connection. And that's actually a good thing.

What the automation did was take all the admin friction away from the stylist. They didn't have to remember to ask. They didn't have to send the text themselves. The system handled that. And because the friction was gone, the conversion rate went up.

AI doesn't replace the human. It removes the friction around the human. The stylist's talent is still what drives the review. AI just makes sure the ask happens every single time.

The Education Gap Is Real

Daniela told me she'd been spending three to four months just seeding her audience on what AI agents even are. What the vocabulary means. How to think about automation.

That felt slow to me at first. But she was right.

In non-tech industries, you can't skip the education phase. People need time to understand the language before they can say yes to the solution.

This is also why workshops work so well for this market. You're not selling software. You're giving people a live experience of what's possible. Once someone sees an AI agent do something in real time that they thought was impossible… they're sold.

What to Do If You Are in One of These Industries

You have more time than the tech crowd, but not unlimited time. The window where you can be the first person in your specific niche doing this is probably 6-18 months wide, depending on your industry.

Start with whatever is closest to revenue for your specific business. If reviews drive clients, start there. If social media drives your pipeline, start there. If proposals are eating your time, start there.

And find someone who can show you what's already working in other industries, then adapt it to yours. You don't need to figure this out from scratch. The playbooks already exist.

The spa industry is behind. Construction is behind. Most of healthcare is behind. But the people in those industries who get moving now won't be behind for long.

Interested in AI automation for your business? Thanh runs hands-on AI workshops in Austin and works directly with business owners on implementation. Learn more here.

I pulled up my screen and showed her something I call a dossier.

It was a Google Doc the AI had generated about her before our call — pulled from her LinkedIn, her organization's website, and a timeline of her nonprofit work. Nothing private. Just public context, organized nicely.

She looked at the screen and said, “Wow, that's so cool.”

The whole thing took 45 seconds to generate.

That was the meeting prep agent. But it's just one piece of a larger system.

What “Digital Chief of Staff” Actually Means

When most people hear “AI assistant,” they picture a chatbot. Something you type questions into and get answers back.

That's not what I'm talking about.

A digital chief of staff is a set of AI agents — each one handling a specific job that used to eat your time. Meeting prep. Follow-up emails. Introduction drafts. Daily briefings. Task creation from spoken commitments.

You still make every decision. You review everything before it goes out. But the legwork — the research, the drafting, the coordination — that all happens in the background.

Think of it as AI digital employees. Not a single magic tool. A small team, each one trained on one job.

The Three Jobs That Changed Everything

Here's the specific setup I use most, and the one I now help clients build.

Job 1: Meeting prep in 45 seconds

Before any first meeting, I get a Google Doc on the person. LinkedIn data, professional history, org context, sometimes a personalized visual. I scan it for 45 seconds before I join the call and I'm up to speed.

Without this, I was either going into calls cold or spending 10-15 minutes manually looking people up beforehand. For someone who has 5-8 meetings a day, that adds up fast.

Job 2: Follow-up emails before I leave the room

The moment a meeting ends, the AI has already drafted the follow-up. It pulls from what was said in the transcript and writes in my voice. I open my inbox, read the draft, make any edits, and hit send.

I built a daily briefing system for a VC friend that does something similar — normally it would take a human assistant 20 hours a week to research every person he's meeting, find old email context, and summarize the latest news on their companies. Now he gets a prep doc every morning and spends his time on strategy instead of inbox archaeology.

Job 3: Introduction emails that used to take 20 minutes

I make a lot of introductions. I care about them being thoughtful — not lazy “you should know each other” emails with no context.

That used to take me 20 minutes per email. I'd look up both people, think about what they have in common, write something personal.

Now I just mention the name. The agent knows both people from my CRM and past transcripts. It drafts the email. I review it. Done.

20 minutes down to about 90 seconds of review.

The Concept Behind It: Agent as Teammate

What makes this work is not the individual tools. It's the mental model.

Most people build AI workflows as one-off automations — “if this happens, do that.” That's fine for simple stuff.

But a chief of staff is not a flowchart. It's someone who understands your priorities, remembers context across conversations, and knows when to act versus when to ask.

That's what I'm aiming to replicate. I think of this as building an agent as a teammate — aware of who you're meeting, what you've committed to, what your calendar looks like, and what tone you use when you write.

The more context the agents have, the better the output. This is why I keep a master context document — a single file that tells the agents who I am, what I care about, how I communicate, and what my current priorities are. Every agent reads it.

What People Save When They Set This Up

The number I hear most often from people who've implemented this: 20 hours a week.

That's specifically the admin work that wraps every meeting — prep research, follow-up writing, intro emails, CRM updates, task logging.

None of that is hard work. It's just work that eats time.

And the interesting thing is, once it's automated, you realize how much of your day used to be spent on stuff that wasn't actually your job. Your job is the thinking, the decisions, the relationships. The prep and the follow-up are infrastructure.

How to Start Building Yours

You don't have to build 14 agents at once. I didn't.

I've been using Lindy since early 2025, and my current system is the result of incremental improvements over months. I built one agent. It worked. I built another. Iterated. The system I have now is not something I sat down and architected in a weekend.

A good starting point:

1. Pick one meeting each week where you wished you'd been more prepared going in
2. Build a prep agent for that meeting type — ask for LinkedIn and org context plus one talking point
3. Run it for two weeks. See how it feels.
4. Then tackle follow-ups.

One agent at a time. That's how you get to a full system without burning out or building something that falls apart.

The goal is not to automate your personality. It's to handle the grunt work so your actual personality has more room to show up.

Want to see a digital chief of staff built live? I run one-day AI workshops where we build your first agents together. Drop me a note if you want details on the next one.

Jacob had one semester left. He was interning at a home builder in the area. And his dad had specifically reached out to me because of something the company told Jacob: if you can help us with AI implementation, you jump to the top of the candidate list.

Not a software company. Not a startup. A home builder.

That conversation stuck with me because I hear versions of it constantly now. The job market is shifting in real time, and most people are still treating AI as optional.

It's not optional anymore.

What the Home Builder Story Actually Shows

Jacob's situation is interesting because it flips the usual narrative. We talk a lot about AI taking jobs. But here's a company in a physical, blue-collar-adjacent industry actively rewarding the employee who can bring AI skills.

Why? Because the home builder had real problems, AI could solve. Proposal tracking. Project status reports. Communication between site teams and office staff. These aren't exotic problems. They're just problems no one had the time to fix.

Jacob's dad made a decision I respect. He didn't want to hire someone to build AI tools for Jacob's employer. He wanted Jacob to learn how to build them himself.

His exact reasoning: “If he learns this, he can be valuable to people.”

That's the right call. And it's something I've been saying in my workshops for two years.

The Difference Between Using AI and Building with AI

There are three levels of how people work with AI. I call them the AI Fluency Levels.

The first level is AI Assisted – you're having conversations with ChatGPT, getting writing help, maybe using it for research. Good start. Most people are here or haven't started at all.

The second level is AI Workflows – you've connected tools together. Emails go to a folder, get summarized by AI, and you review a digest. You've automated something.

The third level is Building Agents – you've created something that runs without you. A system that handles a specific task end-to-end while you focus on something else.

Jacob's employer isn't looking for someone who can chat with GPT-4. They want someone who can get to level two or three. Someone who can look at a 15-page project status report and say: I can build an agent that converts this to a one-page visual every Friday.

That's what creates real value. And that's what moves you to the top of the list.

I've Seen This Story More Than Once

Around the same time I was working with Jacob, I was coaching another student. He'd been trying to get hired at a construction company for six months. No luck.

After five weeks of working with me… about an hour a week… he built a web app from scratch. He went to the job site, asked the workers what slowed them down, and built a project tracking tool that calculated whether timelines were on schedule.

When he reapplied with the app in hand, management reconsidered him.

Not because he had more experience. Because he showed up with a solution to their actual problem.

That's the shift happening right now.

Why Most People Miss This Window

Here's the thing that frustrates me.

If you're on Twitter or LinkedIn, you think everyone is building AI agents and the window is closing. But I talk to actual business owners every week. Small companies, service businesses, construction firms, clinics. Most of their staff barely knows how to write a prompt. The average employee is nowhere close to building an agent.

The window is still wide open.

I ran my first AI workshop in Austin in early 2025. Sold out immediately. My second one: five repeat attendees from the first workshop, just trying to stay current. The hunger for practical AI skills is real, and the supply of people who can actually teach it in a hands-on way… not seminars, not YouTube videos, but real hands-on workshops… is still tiny.

What to Actually Do

If you're early-career or trying to differentiate yourself, here's the simplest frame.

Start with one thing. Not five tools, not an entire automation stack. One problem at work that eats 30 minutes a week. Build something small that solves it. Document before and after.

That single project is now something you can put in front of an employer.

The One Tweak a Week approach works here. Add one more AI capability to your toolkit each week. In three months, you've built 12 small things. Some won't work. Some will. The ones that work become your portfolio.

Jacob's dad was right. Learning to build is different from having it built for you. The skill compounds. Every project teaches you something the next one needs.

Home builders are asking for this. Construction companies are asking for this. Clinics, real estate firms, law offices – they all have the same problems. And they're all starting to notice who walks in the door with real solutions.

That person doesn't have to be a programmer. They just have to be curious enough to start.

Thanh Pham runs AI workshops in Austin and works with companies across industries to implement practical AI systems. If you want to learn hands-on AI skills for your career or business, check out the Productivity Academy or reach out directly.

You get off a call. You made two or three promises. You're onto the next thing, and one of those commitments quietly disappears. Not because you're disorganized. Just because you're human.

I used to handle this with sticky notes, a Todoist list I'd update manually after each call. It kind of worked. But I was still the bottleneck.

About four months ago, I built an agent that removed me from that process entirely.

What the Agent Does

My AI notetaker handles 9+ tasks after every call. The one that changed my day-to-day: it listens for verbal commitments and creates a Todoist task with the right due date automatically.

If I say I'll get you a proposal by Monday on a Wednesday call, I close Zoom, and a task is already in Todoist: Send proposal to [name]. Due: Monday.

The full sequence after each meeting:

1. Meeting gets recorded and transcribed
2. Summary generated
3. Action items pulled out, tasks created in Todoist with due dates
4. CRM updated with notes from the call
5. Draft follow-up email queued for my review

The whole thing runs in the background while I grab coffee.

Why I Built It This Way

There's a principle I come back to: automate the stuff that happens every day first. I call this the 80/20 rule for agent building. Daily loops create compounding ROI. Quarterly workflows just don't move the needle the same way.

Meetings happen every day. Admin after meetings happens every day. That's the loop worth building first.

What Happened With a Client

A few months back I deployed a version of this for a commercial real estate client. His sales team had the universal sales team problem: great at calls, terrible at CRM admin afterward.

Same agent concept. Calls end, the agent transcribes, CRM updates automatically, action items created, follow-up emails drafted. What used to be 30 minutes of post-call admin per rep now happens instantly.

The Email Cleanup Layer

I also layered in a weekly inbox cleanup agent. Every Friday at 9am it goes through my inbox and archives emails older than 7 days. But with rules: anything related to LATT3, Padel Society, or fundraising always gets flagged. Promotional stuff just gets cleared out.

Two layers: real-time triage during the week, deep clean on Fridays. My inbox is finally not a source of stress.

Where to Start

I have 40+ agents running now. When I show my agent library in workshops, people sometimes feel overwhelmed. But I've been building since early 2025, adding one at a time.

You don't start with 40. You start with one. The thing that happens every day and drains a little energy every time. For most people I talk to, that's the meeting notetaker.

If you want to see the setup, drop a comment or reach out. I run workshops in Austin where I walk through this live.

Visit www.asianefficiency.com for more productivity tips and tactics.

Masterclass.com/TPS for at least 15% off any annual membership.

Go to HomeServe.com to find the plan that's right for you. Not available everywhere. Most plans range between $4.99 to $11.99 a month your first year. Terms apply on covered repairs.

Ready to tackle bigger problems? Go to Claude.ai/tps.

Cheat Sheet

• Top 3 Productivity Resources [1:46]
• The shift that changes everything: when AI stops advising and starts executing [5:53]
• Why the Mac Mini is sold out everywhere (and what it unlocks for your workflow) [9:32]
• The uncomfortable test for whether you're actually using AI or just playing with it [14:07]
• The $1 morning briefing that replaces your executive assistant [22:27]
• How to stop your AI from reminding you about things you've already handled [28:39]
• Three things a digital chief of staff does that you probably didn't know were possible [37:29]
• Thanh's personal rule for what AI can do without checking with him first [46:54]
• Four AIs coded and argued for five straight days while Thanh watched a paddle tournament [50:36]

Links

• MasterClass
• HomeServe
• Claude
• 25X Productivity Coaching
• Lindy
• Perplexity Computer
• Wispr Flow
• What’s Possible with AI in 2026: From Flashy Demos to Quiet Leverage (TPS600)
• Obsidian
• Brave Search
• Otter
• Granola
• Mac Studio
• Hermes Agent
• OpenAI Codex
• Qwen 3.5

If you enjoyed this episode, follow the podcast on Apple Podcasts, Spotify, Stitcher, Overcast, Pocket Casts or your favorite podcast player. It’s easy, you’ll get new episodes automatically, and it also helps the show. You can also leave a review!

“Do the people who attend your workshops actually walk away knowing how to use AI? Or do they just feel inspired for a day?”

Fair question. Because most AI training does exactly that — leaves you inspired for a few hours, then back to square one by Monday.

Before I answered, I did something. I pulled up a Google Doc my AI agent had auto-generated before our call. It had Monica's full professional timeline, her LinkedIn data, the CC4C nonprofit's background, and a personalized image of the organization.

She didn't know I'd built it before we spoke.

Her reaction: “Wow, that's so cool.”

And then I said: “That's what you'll be able to build yourself by the end of the day.”

That's the whole philosophy behind my workshops.

The Problem With Most AI Training

Most AI education is passive. You watch a demo. You nod along. You think, “that looks useful.” You go home.

And you do nothing differently.

This isn't because the content is bad. It's because of the format. Watching is not the same as doing. You can watch a thousand hours of cooking videos and still not know how to cook.

I tried online courses for a while. Recorded videos, structured modules. The problem: people don't finish them, and even when they do, transfer is low. They can describe what they learned but can't reproduce it on their own.

The format that actually works is what I call the “let's do this together” model.

I demo it. You try it. You hit a wall. I'm right there to fix it with you, live.

What Hands-On Teaching Looks Like

I had a friend named Jacob come through a hands-on session earlier this year. Zero coding experience. Six weeks later he'd gone from zero to intermediate — setting up Docker, using Claude Code, shipping a production app for construction superintendents.

He told me directly: watching hundreds of YouTube tutorials didn't come close to one session with real feedback. His rate of learning was different. Faster. More durable.

That's not surprising. In the AI Fluency Levels framework I use in my workshops, there are three stages: AI Assisted, AI Workflows, and Building Agents. Most people try to skip straight to agents without mastering the basics. You can't really teach stage three without someone watching you do stage one wrong first.

Catching the mistake live is worth more than a hundred explainer videos.

The 14-Minute Manufacturer Story

Here's a concrete example of what this looks like in practice.

I was working with Gwyneth, a jewelry designer who wanted to transition from fine jewelry to semi-fine. She needed a manufacturer who could do it — but not in China, in a specific geographic area. She'd been searching for two weeks. Getting nowhere.

I showed her how to do it in 14 minutes using AI research workflows. The right combination of prompting, search tools, and structured output.

“Little things like that,” I told Monica, “that you can just learn and discover that would make your life day to day a lot easier.”

That story wouldn't land in a YouTube video the same way it lands when someone watches you do it in real time. In a live setting, Gwyneth gets to watch the exact prompt, see the output, ask questions, and then try it herself. The learning sticks because she did it, not just watched it.

Every Meeting Can Become a Learning Moment

One of the side effects of running a lot of workshops: I've gotten very good at capturing what happens in them.

I use a transcript-first approach. Every meeting, every coaching call, every workshop session gets captured and stored. And then something interesting happens — AI can turn those transcripts into marketing assets, case studies, teaching examples, even blog posts like this one.

This is what I call the story bank concept. Every meeting you have is a potential source of real content. A donor impact story from a nonprofit team meeting. A time-savings win from a one-on-one coaching call. A “wow, that's so cool” moment from a discovery call.

Most people let those moments evaporate. With the right setup, they become a searchable library of real proof points your AI can draw from whenever you're creating content.

Generic AI content uses fabricated stories. Your content can use real ones.

The Difference Between Inspired and Changed

I ended that call with Monica by asking what kind of training her team actually needed.

She said she didn't want them to just understand AI. She wanted them to use it.

That's the dividing line. Seminars create understanding. Workshops create skill.

Understanding fades. Skills don't.

Laurina, who attended a workshop in Virginia Beach, said it best — she was reaching for paper to take notes when she stopped and laughed: “You probably have an agent taking notes right now, don't you?”

Yes. The whole conversation was being automatically transcribed and organized. She left not just knowing that was possible, but knowing how to set up the same thing herself.

That's the format. That's the goal. Not “wow, that's cool.” But “I can do that now.”

Want to actually use AI, not just learn about it? My next full-day workshop in Austin is coming up. Nine to five, hands-on, personalized by your industry. Reply to this post or reach out directly if you're interested.

Thanh runs AI workshops for founders, investors, and business owners in Austin and remotely. 88 NPS in 2025.

Inside the email was a graphic. It laid out everything we had talked about in the meeting, visually summarized, with the key points organized so anyone on their team could scan it in 30 seconds. No bullet-point wall of text. An actual graphic.

They hired me to do their AI strategy the same day.

I want to be clear about something: I was not trying to be impressive. I was not pulling some sales trick. That email was automated. My meeting notetaker agent had already joined the call, listened to everything, generated the summary image, and drafted the follow-up before I even stood up from my desk.

I reviewed it. Hit send.

That was it.

Why the Follow-Up is the Most Underrated Part of Sales

There is a thing I tell people in my workshops: the faster your follow-up, the more business you close. It sounds obvious. And yet most people send follow-up emails hours later, or the next day, or sometimes not at all.

AI does not procrastinate.

I built a workflow where my notetaker agent joins every Zoom or in-person call I take. When the meeting ends, it:

1. Transcribes everything
2. Pulls out key decisions, action items, and next steps
3. Generates a visual meeting summary (an infographic-style graphic)
4. Drafts a follow-up email in my voice, referencing specific things from the conversation

Within minutes, the draft is in my inbox. I review it, make any tweaks, and send.

The whole thing takes me about two minutes.

And the people on the other side of that email? They are blown away. Not because I am special. Because almost nobody does this. You send a personalized follow-up with a polished visual summary within five minutes of hanging up, and you stand out from literally everyone else they talked to that week.

The Super Agent Behind the Scenes

Here is what I actually built, and why it matters for anyone thinking about AI tools.

I have what I call a super agent. It is a Lindy workflow that connects three things ChatGPT cannot connect on its own: my email, my calendar, and my Google Drive.

When I need a project update on a key client relationship, I ask this agent to catch me up. It searches the last 30 days of email with that person, cross-references calendar events, matches the two together, and creates a Google Doc with a full status summary. Then it links that doc to my upcoming calendar event with them so I have it ready before the meeting starts.

Three steps. One request. Done.

I showed this live in my Two Hour Workday course and the reaction was basically: “Wait, that is it?”

Yeah. That is it. The power is not in complexity. It is in connecting things that most people still do manually, or do not do at all.

The 42-Cent Agent (My Favorite Example)

I want to give you a smaller example too, because I think people get overwhelmed when they hear “AI agents.”

I have been car-free in Austin for over 11 years. I use Uber for almost everything. And I constantly forget to add addresses to my calendar events, which means when I am about to leave I am scrambling to look up where I am going.

So I built a tiny agent. It wakes up every Monday, Wednesday, and Friday at 8am. It scans my calendar for the next five days. Any event that looks like it involves a physical location but is missing an address, it Googles the address and adds it to the calendar event.

Cost: 42 cents a week.

That is what I mean by start small. I am not saving 20 hours a week with this one. But it removes a small, consistent friction that was annoying me every single week. And it runs whether I think about it or not.

This is the framework I use when thinking about what to build: the 80-20 rule applied to automation. Do not build the big impressive agent first. Build the one that handles a thing you do or deal with every day. Those are the ones that compound.

Where Most People Get Stuck

The question I hear most often in my workshops is not “how do I build this?” It is “what should I build first?”

And the answer is almost always the same: what is the thing that happens every day or every week that you wish just happened on its own?

For a lot of people, that is meeting follow-ups. Or email triage. Or preparing for the next day calls.

Start there. Build something small that works reliably. Use it. See what it saves you.

Then expand.

That is the whole philosophy behind start small, iterate. Life gets better one agent at a time. You do not need a 20-agent system on day one. You need one agent that you actually use and trust. That trust builds. And then you add the next one.

The meeting summary infographic that closed a real estate deal? That started as a simple notetaker experiment. I did not know it would win me business. I just wanted to stop spending 30 minutes writing follow-up emails after every call.

The side effect was getting hired on the spot.

Try This This Week

If you want to start somewhere:

1. Look at the last 5 meetings you had. How many got a same-day follow-up? How many got a follow-up that included a summary the other person could share?
2. If the answer is “not many” — that is your first agent to build.
3. Tools like Lindy can connect your notetaker to your email and have drafts ready within minutes of a call ending. Start with that workflow.

One agent. Daily use. See what happens.

Thanh runs AI workshops and consulting for entrepreneurs and executives. If you want to see these workflows live, check out the next workshop or reach out directly.

Midway through the call, she said something that's stuck with me.

“I think I need people who are NOT seasoned. Because they don't come in with those preconceived notions.”

She wasn't complaining. She was having a genuine insight in real time.

Her senior bookkeepers… the ones with 10 or 15 years under their belts… kept dragging their feet with the AI tools. Not openly hostile, just passive. “I like checking it myself.” “What if it misses something?” They'd give it a try, notice one small thing it got wrong, and use that as proof it wasn't ready.

Her junior staff? Two days. Done. Using it, trusting it, moving on.

The problem with expertise

Here's what I think is happening.

When you get really good at something, you build systems inside your head. Not formal systems. Just… the way you do things. The order you check stuff in. The mental shortcuts you've developed. The feel for when something's off.

That's valuable. My client's senior bookkeeper caught a $65K error last tax season that three other team members had missed. The client's tax bill had been calculated at $90,000. Should have been $25,000. The preparer missed it. The reviewer missed it. A second reviewer missed it. She caught it in about 30 seconds because she knew the client well enough to know the number was wrong.

That's expertise. Real expertise. Hard to replicate.

But the same pattern that makes expertise powerful also makes it resistant. When AI enters the picture, it doesn't just change the tools. It changes where the skill lives. And that feels threatening to someone who's spent years building competence in a specific way.

The junior person? They haven't built that identity yet. The new tool isn't replacing anything. It's just… how they learn to do the job.

What this actually means for hiring and onboarding

I'm not saying you should replace experienced staff with fresh graduates and call it an AI strategy. That's a terrible idea.

But I do think the client's insight points to something real: the way you onboard people into AI tools matters a lot, and it probably needs to be different for different team members.

For newer staff, get out of the way. Give them the tool, a basic orientation, and let them explore. They'll figure out where it helps fastest.

For experienced staff, you have to do the opposite. You have to explicitly honor what they already know. The framing isn't “here's a tool that will make your job easier.” It's “you have expertise this tool doesn't have. Here's how we use the tool for the repetitive stuff so your expertise has more space to do what only you can do.”

That framing matters. Because it's also true.

Before you automate: map it first

One thing I always tell my consulting clients is that AI adoption problems are usually workflow problems in disguise.

Before we build any agents or set up any automations, I want to do a process mapping session first. Literally get a visual of every step someone takes in their workflow. What happens when an email comes in? Who touches it? What decision gets made? Where does it go next?

Once you can see the whole workflow on a whiteboard or a screen, it becomes obvious where automation helps and where human judgment is non-negotiable. And that map is also the tool you use to show your experienced team: “Here are the parts we're automating. Here's the part where you're still the expert.”

That conversation goes much better when people can see it.

The real question isn't “will they adopt AI”

It's “what are they afraid of losing?”

A front desk person who's built a career on being the hub of information at a clinic… she's afraid AI makes her redundant. A bookkeeper who's proud of never missing an error… she's afraid AI gets the credit for the work she's been doing for 15 years.

Those fears are reasonable. They need to be addressed directly, not assumed away.

The clients I've seen do this well are the ones who bring their team into the process early. Not to teach them how to use the tool. To ask them: what are the parts of this job you hate? What takes the most time but adds the least satisfaction? Let's start there.

When you automate the stuff they hate, it stops feeling like a threat. It starts feeling like a gift.

One thing to try

If you're rolling out AI tools at your company and hitting resistance, try this before anything else.

Talk to the resistors one on one. Not about the tool. About their job. Ask them what parts of their week feel like a waste of their skill. Ask what they'd do with that time if the repetitive stuff disappeared.

You'll probably find out the problem isn't the tool at all.

And you'll learn something useful about your own workflow in the process.

Thanh Pham runs AI workshops for business teams and consults on automation strategy. If you want help mapping your team's workflows before rolling out AI tools, the workshop is a good place to start.

73 hours. Across over 1,000 tasks.

The number that surprised me most wasn't the total. It was the breakdown. One single agent — my email inbox manager — had done 198 tasks and saved me almost 18 hours in that one week.

Just email. Nothing fancy.

I share that screenshot in workshops now. People stare at it for a second and then ask: “What kind of complicated system did you build for that?”

And I have to tell them… it was just the inbox.

Why Email Is the Right Place to Start

Most people I talk to want their first AI agent to be impressive. Something that monitors multiple data sources, synthesizes information, and outputs strategic recommendations. Something they can show people at a conference.

But the 80-20 of agent building says: start with frequency, not sophistication.

What you do every single day is worth more to automate than something you do once a quarter. Email is the thing most knowledge workers touch the most, groan about the most, and almost never think to automate.

I've spent a lot of time teaching this stuff, and the pattern is consistent. People skip email because it feels too basic. And then they wonder why their fancy automation isn't saving them much time.

What the Email Agent Actually Does

Here's what mine handles:

1. Reads every new email as it arrives
2. Categorizes it: needs action, FYI only, can delete, needs follow-up later
3. Drafts a reply in my voice if a response is needed
4. Flags anything that actually requires my judgment

What I do: open the review queue, check the drafts, edit if needed, click send. Maybe 20 minutes total. For a day of email.

The first time I taught this in a workshop, someone raised her hand and said: “But isn't it scary to have AI handling your email?”

I get that reaction. A lot of people are worried the agent will send something on their behalf without them approving it. Mine never does that. It drafts. I review. That's the human-in-the-loop setup I always recommend for anything that goes out with your name on it.

Once you trust the drafts — once you see they actually sound like you — that's when you start moving faster.

The Telegram Piece Nobody Expects

Here's the part that surprised even me.

I interface with my email agent through Telegram. It's just a messaging app I already used, so adding this layer felt natural.

Walking to a meeting, I'll dictate: “Draft an email to Evan introducing Lauren Goldstein as a potential guest for the members club. She'd be a great fit.” Thirty seconds later there's a draft in my inbox. Ready to review.

I didn't sit down. Didn't open my laptop. Just talked.

Combine that with Whisperflow — which uses AI to understand intent, not just transcribe words literally — and dictating emails while walking between things became genuinely faster than typing at a desk.

That's the shift I didn't fully expect. It's not just that email takes less time. It's that email gets handled at times of day when it previously couldn't happen at all.

You Don't Have to Be Technical

This one I want to say clearly.

I am not a developer. I can't write code. A year ago I had no idea how to build any kind of automated workflow.

The barrier has dropped that fast.

Lindy, the tool I use for most of this, has a visual interface and a chat editor. You describe what you want in plain language and it builds the workflow. The most common fear I hear from workshop participants — “I'm not technical enough for this” — disappears pretty quickly once they're actually in the tool.

I've watched accountants, salon owners, and real estate agents build working agents in a single afternoon. None of them had a technical background.

What it does take is willingness to start small. Most people want to build everything at once. The agents that actually get used are the ones that do one thing reliably.

Email is that one thing for a lot of people.

The Progression

If you want to get here, the path is pretty simple.

Start with AI-assisted work: use ChatGPT or Claude to help you draft individual emails. Just that. Get comfortable with what good outputs look like in your voice.

Then build a simple workflow: something that drafts replies to a specific category of email automatically. Maybe just client inquiries. Or meeting requests. Pick one.

Once that's running and you trust it, expand.

This is what I call the AI fluency progression. Assisted, then automated, then agents. You build trust at each level before moving to the next. Skipping to agents without the earlier stages usually means the first thing you build doesn't get used because you don't trust it yet.

The 18 hours didn't happen overnight. They built up over a few months of incremental improvements.

But at some point you add up the weekly numbers and the total is 73 hours in a week, and you go back and look at which agent is doing the heavy lifting.

Almost always, it's email.

Try This Week

Pick the email category you spend the most time on. For most people that's client questions, scheduling requests, or follow-ups after calls.

Spend 30 minutes building one draft agent for just that category.

Don't try to make it perfect. Just see if it drafts something you'd actually send with a small edit.

That's the whole experiment. And if it works, that one agent will save you more time than anything else you build for a while.

Want to learn how to build this for yourself? The Productivity Academy has workshops on AI agent building, including the exact setup I use for the email inbox manager.

“It still feels weird. Like the wifi is broken or something.”

She meant it as a compliment. I think.

What “Wifi Is Broken” Actually Means

Before the agent, Amanda’s inbox was running in the background of everything. Not just open in a tab. Actually running. The mental check-ins between client files. The “let me just see if anything came in” that breaks a 45-minute focus block into two 20-minute ones.

Tax season is brutal for CPAs. The work itself isn’t the problem. It’s that the work happens inside a constant buzz of emails, CPE requirements, team coordination, and client questions that feel urgent but usually aren’t.

The Lindy agent I built for her does a few things. It auto-categorizes CPE emails and adds the trainings to her calendar. It forwards team-related emails to the right people. It handles inbox routing for the stuff that can be handled without her.

First week, she did three hours of uninterrupted deep work on tax returns.

Three hours. During tax season.

That “wifi is broken” feeling? That’s the inbox being silent for the first time in years. Not because nothing came in. Because the agent handled it.

The Part I Didn’t Expect

A few weeks later, Amanda called me. She wanted to tell me something.

She’d started building her own thing.

A custom GPT that could read a tax return and flag checklist errors before they hit her desk. She’d spent about 90 minutes iterating with ChatGPT. Got a working version, even if it wasn’t perfect yet.

“I actually tried to build a custom GPT that could read a tax return and analyze it for certain points,” she told me. “It asked me to drop the return in there, redacted, and I was working on building it. I was able to do that because Lindy is saving me time. I actually have more time to iterate with GPT.”

She built a second AI tool. On her own. Without me.

That’s not something I planned for when we started. It’s what happens when the first automation actually works.

The Compounding Effect Nobody Talks About

Most people think about AI automation as a time savings calculation. If this takes me 30 minutes every day and the agent can do it, I get 30 minutes back. That math is real.

But there’s a second effect that’s harder to quantify: when you get your attention back, you start seeing differently.

Amanda wasn’t just saving time. She was getting back the mental bandwidth to ask “what else is possible?” And she had enough slack in her day to actually experiment.

This is what I mean when I say start small and build from there. The email automation wasn’t the end goal. It was the foundation that made the next thing possible.

How to Actually Build This

Here’s a technique I use with clients that almost nobody talks about.

If you want to build an AI agent that replicates your own expertise — like Amanda’s tax review GPT — screen record yourself doing the work.

Not just capturing the screen. Narrating out loud. Talk through exactly what you’re looking at, what you’re noticing, what flags you’re catching and why.

Do it two or three times on different cases.

Then give the recordings and the transcripts to ChatGPT. Tell it: based on how I think through this work, help me codify this into a system. A checklist. A custom GPT. A review process.

Your own narration becomes training data. The AI learns your judgment from watching you work, not from a prompt you wrote trying to explain it.

This works for tax reviews. It works for sales calls. It works for any expertise that’s hard to put into words because you’ve been doing it so long it’s automatic.

What the $65K Error Taught Her About AI and Human Judgment

One more thing Amanda told me, and it stuck with me.

A client came in saying they owed $90,000 in taxes. Should have been around $25,000. Before she even pulled up the return, she knew something was wrong. Her gut immediately went to three possible places it could be.

She was right. A bookkeeping error that had passed through three different team members. Nobody caught it.

AI can handle the repetitive, rules-based work. The sorting, the routing, the formatting. But that pattern recognition she has after years of seeing thousands of returns? That’s not automatable yet. And it’s worth protecting.

That’s the right framing for AI in a professional services business. Not: AI replaces the expertise. It: AI clears the path so the expertise can actually be used.

When Amanda isn’t sorting CPE emails, she’s catching $65,000 errors.

That’s the trade worth making.

Curious what it looks like to set up an AI inbox manager for your own practice? I run one-day workshops where we build these systems live. Or reach out directly if you’d like to explore a custom setup.

Now? If offline mode is listed as a feature, that's actually a yellow flag. It means the team is still treating it as optional.

In 2026, cross-platform availability and offline support are table stakes. Not selling points. Not differentiators. The baseline.

And if an app doesn't clear that bar, I don't spend much time on it.

The Client With Three Note-Taking Apps

A while back, I was working with a client on a productivity overhaul. He was a sharp guy — running a solid business, staying on top of a lot.

But he had a note-taking problem.

He was using Evernote, Apple Notes, and Notion — all simultaneously. I asked him which one he used when he needed to save something.

“Whichever one is open,” he said.

And to find something later? He'd try to remember where he'd put it. Or just search all three.

That's not a system. That's chaos with extra tabs.

But here's the thing — this wasn't a him problem. It was a tool problem. When apps work poorly offline or only on certain devices, people naturally fragment across multiple tools. You open what's convenient. You save where it loads fastest. And over time you end up with your notes spread across three apps and no real home for any of them.

The fix wasn't a new habit. It was picking one tool that worked everywhere.

The Notion Lesson

Notion is a great example of this playing out in real time.

For years, Notion's offline mode was rough. Not non-existent, but rough. You'd open it on a flight or in a spotty wifi zone and get inconsistent behavior. Notes that wouldn't load. Edits that wouldn't save.

It got much better eventually. But for a long time, that was a real problem for anyone who relied on it as their main system.

And I felt it firsthand. I was traveling pretty frequently during that stretch, and every flight was a reminder that I couldn't fully trust my notes system without a wifi connection.

That kind of friction is sneaky. It doesn't break your workflow all at once. It just adds a little bit of doubt every time you open the app. You start keeping a backup. Then the backup becomes the real system. And now you're back to two apps.

My Checklist Before Committing to Any New Tool

Over the years I've developed a short filter for any new productivity app. It's not comprehensive — it's just the stuff that saves me from bad decisions.

Does it work offline?

Not “kind of offline.” Not “you can view cached content offline.” Actually offline — you can create, edit, and save without a connection, and it syncs when you reconnect.

Is it cross-platform?

I work across a Mac, iPhone, and sometimes an iPad. If the app lives only on one of those devices, it's not a productivity app for me. It's a partial solution. And partial solutions create the fragmentation problem.

Would I commit to this for at least 2 years?

This one's underrated. App switching costs are brutal. When you move notes, you have to migrate content, rebuild your system, and relearn the shortcuts. If you're switching every 6-12 months, you're spending a huge chunk of your time and mental energy just on tool maintenance — not on actual work.

I follow a rough 90-day forced trial rule for any new tool I'm evaluating seriously. The first few weeks feel clunky no matter what. It takes about 90 days to really know if something will stick and serve you well long-term. If it's still not clicking at 90 days, that's useful data.

Is it at least 10x better than what I already have?

Switching costs are real. If a new tool is only marginally better than what you're using now, it's probably not worth the disruption. The bar for switching should be high — not “this is slightly more organized” but “this changes how I work in a meaningful way.”

What Mac-Only Apps Get Right (and Wrong)

There's a nuance here worth mentioning.

If an app is Mac-only, I can work with that — because I'm mostly on my Mac anyway. Being device-limited is a smaller problem than being offline-limited.

But an app that's online-only? That's a deal breaker for me. Because the moment I'm on a flight, in a remote workshop location, or anywhere with spotty connection, I need my tools to keep working.

The best apps treat offline as part of the core experience, not an afterthought. Todoist does this well. Obsidian does this really well — it's actually fully local by default.

The Real Productivity Win Here

The point isn't really about offline mode. The point is about trust.

When you trust your tools to work wherever you are, you can stop managing your tools and start using them. You don't keep a backup system “just in case.” You don't hesitate before the flight to make sure everything synced. You just… work.

That mental overhead adds up. Every tool decision that adds friction or doubt costs you more than the tool is worth.

So before you download the next productivity app that your colleague recommended or saw on Twitter — run it through the basics first:

1. Does it work offline?
2. Does it work on every device you actually use?
3. Could you see yourself using this in two years?

If those check out, then it's worth evaluating the features. If they don't, move on fast.

There's no shortage of great tools. But there's a real shortage of tools that fit your actual life.

Want to build a cleaner, more sustainable productivity system? Check out the weekly review — it's a good place to start.

Good idea. But I'd just been running it on Claude because… that's what I'd been using. I'd never actually tested whether it was the right choice.

Then one week, I got curious.

The Test

I ran the exact same workflow three times. Same prompt, same input data — a full week of meeting transcripts. The only variable was the model:

• Claude Sonnet 4.5 (cost: 214 credits)
• ChatGPT 5.2 (cost: 217 credits)
• Gemini 3.0 (cost: 274 credits)

Then I read all three outputs against what I actually knew about my week.

What Happened

ChatGPT came in last. Not slightly behind — genuinely last. It skipped things that had clearly happened. Missed patterns that were right there in the text. I don't know if it was the context window, the summarization behavior, or something else, but the output was noticeably thinner than the other two.

Claude was the runner-up. Solid, complete, nothing obviously missing. But nothing surprising either. It gave me back a clean version of what I already knew.

Gemini 3.0 won. And it wasn't really close.

Gemini surfaced stuff I would have missed on my own. Connections between meetings from different weeks. A theme that had been building for about six weeks that I hadn't consciously noticed. Specific phrasing patterns from client calls that kept repeating.

That's what I was building this workflow to catch. And only one model actually delivered it.

Why This Matters More Than You Think

Here's the thing. I've been teaching AI productivity for a few years now, and one of the patterns I see constantly is what I call model loyalty. People start with one AI tool, get comfortable, and just… stay there.

It's like finding a restaurant you like and never eating anywhere else.

The truth is, each model has different strengths. I think of it this way:

• ChatGPT is the daily driver. The reliable workhorse. Great for general tasks, quick answers, and anything where you need something done fast.
• Claude is the speedboat. Fast, precise, excellent at technical reasoning, writing, and nuanced analysis.
• Gemini is built on top of Google's data. It handles long documents and large amounts of text better than the others, and it surfaces patterns across big datasets in a way that genuinely surprises me.

This is what I call being multi-tool native — you don't pick one AI and stay loyal to it. You match the tool to the job.

The best AI users I know all do this. They route work to wherever it'll get done best. Technical reasoning goes to Claude. Image analysis goes to Gemini. Quick daily-driver tasks go to ChatGPT. Recurring workflows that need integrations go to Lindy.

The Wrong Assumption

When I started testing, I assumed ChatGPT would do fine. It's the most popular, the most talked about, the one most people default to. I figured it would at least be competitive.

I was wrong.

And this is the bigger lesson: most people's assumptions about which AI is best are based on reputation, not testing. They go with the one they heard about first, or the one that feels most familiar, or the one their tech-savvy friend recommended two years ago.

That's a terrible way to build a workflow.

The models are improving and changing constantly. The right call last year might not be the right call today. What works for one task category might fail on another.

The only way to know is to test.

How to Run Your Own Test

You don't need to do anything complicated here. Pick one task you use AI for every week. Something recurring. Something where you have a sense of what good output looks like.

Then run it through two different models side by side this week.

That's it. Just compare the outputs. Don't add any other variables.

You'll get a useful data point. And you'll probably discover something that surprises you — the way my Gemini result surprised me.

From there, you can start routing intentionally. When you have a long document to synthesize, you know which model to reach for. When you need precise technical writing, you know where to go.

That's how you actually get better results from AI… not by finding the one best model, but by knowing when to use each one.

One More Thing

I still run that weekly synthesizer on Gemini. Every Sunday, it reads the week's transcripts, and every Sunday I read something I didn't already know.

I'm spending 274 credits instead of 214. The extra 60 credits is maybe $0.10.

Best $0.10 I spend all week.

Want to learn how to build workflows like this? The Productivity Academy has workshops that cover AI agent design, tool selection, and getting real results from automation. Check it out.

That's not a review. That's recency bias with a journal.

Your brain is not a good archive. It's excellent at surfing the most recent weeks, but it drops most of what happened in February, March, or June. By the time January rolls around, you're basically reviewing a highlight reel of the last 30 days… and calling it a year.

This is why the same goals show up year after year. Not because people are lazy. Because they're planning from memory instead of from data.

Here's the framework we use at Asian Efficiency. Three phases: reflect, synthesize, design. In that order. The review has to come before the goals.

Phase 1: Reflect — With Actual Data

Don't start with journaling. Start with data.

Pull up five sources:

1. Your calendar — scroll through the whole year, month by month
2. Your photos — surprisingly powerful memory trigger
3. Your journal entries (if you keep one)
4. Your credit card statements — where your money actually went
5. Your social media or email history

The calendar and photos are the big ones. Once you scroll through, you start remembering things your memory had buried. A workshop you ran in March. A rough patch in May. A trip that energized you in September.

I had a client once who said fitness was her top priority going into the year. When we looked at her calendar and her spending, fitness barely showed up. A few gym visits scattered through spring. Nothing from July through October. The data told a different story than her memory… and than her stated values.

That's why the data pull matters. You get to see your year as it actually happened, not as you'd like to remember it.

Spend 30-45 minutes on this phase. Go slowly. Take notes as memories surface.

Phase 2: Synthesize — Wins, Lessons, Patterns

Once you have your raw data, the next job is to make sense of it. Organize what you found into three buckets:

Wins. Real ones, not planned ones. What actually happened that you're proud of? What worked? What results showed up that you didn't expect?

Lessons. What went wrong? What did you try that failed? Where did you overcommit, underestimate, or just get it wrong? No self-flagellation here — just clear-eyed honesty.

Patterns. This is the most useful bucket. What kept repeating? What did you keep putting off? Where did the same problem show up in different disguises?

I noticed one year that I kept deferring admin work. Month after month, it showed up in my review as something I “needed to deal with.” I'd tell myself I'd get to it next week. Then next month. Like ignoring a check engine light until the car breaks down. The annual review made that pattern visible in a way that weekly reviews hadn't.

Patterns are where the real insight lives. If something shows up twice in a year, it's a data point. If it shows up every quarter, it's a pattern that needs a system — not just more willpower.

Phase 3: Design — Build the Next Year From Evidence

Now, finally, you plan.

But you're not planning from wishful thinking. You're planning from evidence. You know what actually worked. You know what you kept avoiding. You know what energized you and what drained you.

This changes the quality of your goals.

Instead of “I want to get fit this year” (same goal as last year), you might notice: “I consistently work out when it's scheduled on Tuesday and Thursday mornings, but I never do it when I leave it open.” So the plan isn't just a goal. It's a system. Tuesday and Thursday mornings, blocked, protected.

Instead of “I want to grow my business,” you might notice: “Every time I ran a live workshop, revenue jumped. Every time I tried to create async content, I procrastinated for months.” So the plan prioritizes live events and deemphasizes solo recording projects.

Design based on who you actually are, not who you're trying to become. The data from phases 1 and 2 shows you both.

The Weekly Review Is the Year-Round Version

The annual review is a once-a-year deep dive. But you don't have to wait until December to do this.

The weekly review runs the same underlying logic on a shorter cycle. Every week, you look back at what happened, capture patterns, and set up the next week intentionally.

Most AE frameworks compound through the weekly review. If you only do the annual review, you're course-correcting once a year. If you also do the weekly review, you're course-correcting every week. The annual review becomes much more useful because you've been tracking patterns all year.

The Order Matters

Most people skip to phase 3. They start with goals in January without doing any reflecting or synthesizing first. That's why the same goals repeat.

Reflect first. Let the data surface what actually happened. Synthesize second. Find the patterns and lessons. Design third. Build a plan from what you now know.

Takes a few hours total. Once a year. Worth more than any amount of goal-setting workshops.

If you're doing a quarterly review right now — same process, shorter time window. Pull your Q1 calendar before you write a single Q2 goal.

The data is always smarter than your memory.

Not because I didn't want to watch them. I genuinely did. These were channels I'd deliberately subscribed to… people I respected, analysts I wanted to follow, creators teaching things I needed to know.

But 20-minute videos compete with everything else in a day. So most of them just… sat there.

Then I built a fix.

The RSS Agent Setup

Every YouTube channel has a public RSS feed. Most people don't know this, but it's been there for years. The feed updates every time a new video goes live.

My setup in Lindy monitors 20 channels via their RSS feeds. When a new video drops, the agent wakes up, pulls the transcript from the video, generates a plain-text summary, and sends it to a dedicated Slack channel I call “Learning.”

No link to watch. No “save for later.” Just the summary, ready to read in about 90 seconds.

The information density is actually close to the same. Maybe 80-90% of what I'd get from watching. For most content, that's fine. If something is genuinely worth the full 20 minutes, I can still watch it. But that's now a deliberate choice, not a default.

I went from watching maybe 3 videos a month from those 20 channels to reading every single one.

What This Looked Like in Practice

Last November, I walked one of my coaching clients through this setup. Ilias is a structural engineer who also runs an investing business on the side. He was trying to follow 12 research-focused channels… quarterly earnings breakdowns, sector analysis, market commentary. The kind of content where you genuinely need to stay current.

He'd been saving videos to watch on weekends. Watching maybe 4 or 5. Feeling perpetually behind on the rest.

We built his version of this agent in about an hour. Set it up for his 12 channels, configured the summaries to focus on key data points and analyst takes rather than general content, and pointed it to a private Slack channel.

Two weeks later he told me the part that surprised him wasn't the time saved. It was that he actually followed through now.

Because the friction was gone. The summary was just there. Short. Readable. No decision to make about whether to “find time to watch.”

That's the hidden cost of “watch later” lists. Every item on that list represents a small mental tax every time you open YouTube and see it sitting there, unkept.

Two Agents Are Better Than One

The YouTube RSS setup is a good example of what I call the 80/20 agent building approach: automate the high-frequency pain first.

But there's a variation of this I use for research that goes a step further.

For the Asian Efficiency newsletter, I run two separate agents. The first one is the researcher. It runs three times a week — Monday, Wednesday, Friday — using Perplexity to search across several topics, then stores links and summaries in a Google Sheet.

The second agent is the writer. It runs every Tuesday at 5am, pulls from that accumulated sheet, and generates a newsletter draft.

One agent does research. One agent writes. They don't even run at the same time.

Before I built this, I was spending 4-5 hours every week manually doing that research online. Typing in search terms, opening tabs, taking notes, synthesizing it all. It was one of those tasks I always put off until Sunday night because I dreaded it.

Now the Google Sheet has over 1,000 rows of accumulated research. The writer agent has rich material to pull from every single week.

This is the compound effect of automation. The value isn't just in the time saved per task. It's that the work keeps happening whether I'm paying attention or not.

The Format Problem (Not a Time Problem)

Here's the reframe that I think is worth sitting with:

If you have a “watch later” list, you don't have a time problem. You have a format problem.

Video is a great format for some things. For a tutorial where you need to see someone's hands. For a long-form conversation where tone and context matter. For something genuinely entertaining.

But for most informational content… a summary works better. It's faster. It's searchable. You can skim it.

The reason we default to video isn't because it's the best format for learning. It's because that's what the platform serves you. YouTube is optimized for watch time, not for your retention.

There's a version of this I've taken even further. I set up a personal daily podcast from my Instapaper reading backlog. Every morning, a bot pulls my latest saved article, converts it to audio with my voice preferences, and uploads it to a private RSS feed. I listen to it on my walk or during my commute.

Same information I was “going to read.” Just in a format I'll actually use.

That's the pattern. Take the information you care about and change the container.

How to Start Building This

The YouTube RSS agent setup is one of the better starter automations I recommend because it has a few qualities that make automation stick:

• High frequency (new videos every week from channels you care about)
• Clear output (a summary you can read in 90 seconds)
• Low risk (no external actions, nothing breaks if it misses one)

If you're in Lindy, you can build the basic version in an afternoon. Set the RSS trigger, pull the transcript, write a prompt asking for a plain-text summary with 3-5 key points, and route it to Slack or email.

If you want to get fancy, add a prompt layer that filters by topic. So if you follow a channel that covers both things you care about and things you don't, the agent can flag only the summaries relevant to you.

Start simple. One channel. See how it goes. Then scale.

That's the whole philosophy. Life gets better one agent at a time.

Want to learn how to build setups like this? Check out the Productivity Academy — or come to one of our AI workshops in Austin.

The result surprised me.

Two hours. Most days, two hours of real focused work gets me 80 to 100% of what I used to accomplish in a full workday. Not because I'm working faster or burning myself out in those two hours. Because the other six hours were never really work to begin with.

Let me explain what I mean.

The Problem With Busy

I've been studying productivity for over a decade through Asian Efficiency. And one of the core ideas we keep coming back to is something I call real work vs. fake work.

Real work moves your actual goals forward. It requires your brain, your judgment, and your relationships. Fake work creates the feeling of progress without the actual progress.

Most people's days are 80% fake work.

Email responses. Meeting prep. Status updates. Scheduling back-and-forth. Inbox triage. These things feel necessary. And some of them are. But most of them don't require you specifically. They require someone to do the task.

That distinction matters a lot.

What Changed When I Added AI Agents

Over the past year I've been building what I call a Digital Chief of Staff – a suite of AI agents running in the background of my day.

One agent reads my email and pre-drafts responses before I open my inbox. I review the drafts, make small edits, send. Five minutes instead of forty-five.

Another agent preps a context document before any meeting – who I'm meeting with, what we talked about last time, what decisions are pending, what I need to bring up. I used to do this manually. Now it happens automatically.

A third agent processes every call transcript and pulls out action items, follow-ups, and key decisions. I don't have to re-listen to recordings or build my own notes.

Together, these agents handle probably four to five hours of what used to fill my workday.

And here's the thing: I didn't lose any output. If anything, my output improved. Because when I sit down for my actual two hours of focused work, I'm not mentally cluttered. The admin is done. The prep is done. I'm just working.

The TEA Framework Helps Explain This

At Asian Efficiency we think about productivity through three currencies: time, energy, and attention.

Most productivity advice focuses on time. But energy and attention are often the real bottlenecks.

When I was spending six hours a day on admin, I had the time to do real work in theory. But I didn't have the energy or attention. By the afternoon I was drained. My inbox had already absorbed my best mental hours.

AI agents solved an energy and attention problem more than a time problem.

By offloading the low-judgment tasks to agents, my two focused hours land in the morning when my energy is highest and my attention is sharpest. The agents run in the background while I sleep or exercise. By 9am, the admin is already handled.

This is why I use 9 to 11am as my deep work window, and I protect it hard. Those are the hours that produce 80% of my value. Everything else is support.

This Is Not About Working Less

I want to be clear about something. The Two Hour Workday is not a shortcut. It's not about doing less.

It's about getting the right two hours. Uninterrupted. Uncluttered. Actually focused.

Most people are technically at work for eight to ten hours a day. But how many of those hours involve real thinking? Real decisions? Real creative output?

When I watched an AI agent work through my backlog of emails one night while I was asleep and had 47 drafts ready for me when I woke up, my first reaction was a mix of amazement and embarrassment. Amazement because it worked so well. Embarrassment because I realized how much of my own mental energy I'd been spending on tasks that a well-configured AI could do for forty-seven cents.

That moment changed how I think about what working means.

Where to Start

If you want to try this, start small.

Pick one task you do every day that's mostly repetitive. Email responses. Meeting recaps. Scheduling. Research summaries. Pick just one.

Build or configure an agent to handle it. Give it a week. See how much mental space comes back.

Then build the next one.

The agents compound. Each one you build frees up more attention for the real work. At some point you hit a threshold where you realize your best two hours are more valuable than your previous eight, because they're actually yours.

That's the Two Hour Workday. And once you've lived it, it's hard to go back.

Interested in building your own AI agent stack? I teach this in my one-day AI workshops, personalized by industry. Reply to this post or reach out directly if you want to know more.

She'd been building AI agents for her business. The first few weeks went great. Then things started going sideways. Her email-tagging agent was misclassifying threads. Her meeting prep agent was pulling irrelevant information. She kept tweaking her prompts, but nothing helped.

She thought she needed better prompts.

She didn't. She had a context problem.

The Shift Most People Are Missing

For a while, “prompt engineering” was a real skill. Early models needed precise instructions. Specific syntax. Careful phrasing. You had to talk to AI like it was fragile.

But the models got a lot smarter. And now, most of the old prompting tricks are irrelevant. You can just… talk to the AI naturally. Explain what you need like you'd explain it to a smart colleague.

What matters now is context engineering.

Context engineering is simpler than it sounds. It just means giving the AI enough specific details to do the job well.

Here's the clearest way I know to explain it.

If I type: “Write a newsletter on AI agents,” I'll get a generic, forgettable newsletter that could've been written by anyone.

But if I type: “I'm Thanh from Asian Efficiency. This newsletter goes out to about 80,000 people. Most of them work in corporate America or run a small business. They're relatively new to AI but curious about it. Write a newsletter on AI agents” — now the agent knows exactly who it's writing for. The output is completely different.

Same request. Completely different context. Night and day results.

That's it. No magic words, no special formatting, no secret syntax. Just the right details.

The Context Window Mistake

Here's where I see people go wrong once they understand this idea.

They swing too far in the other direction.

They open ChatGPT, load in 300 documents, paste their entire email inbox, and then ask one small question. Then they wonder why the answer is vague or wrong.

The AI can only hold so much in memory at once. When the context window gets too full, it doesn't crash — it just starts skimming. Accuracy drops. The model starts guessing.

I use a sheet of paper analogy with my workshop students. When a conversation fits on one sheet of paper, the AI reads every word carefully. When you give it a stack of paper, it starts skimming like a student the night before an exam.

My student's problem wasn't her prompts at all. She'd loaded her agents with massive knowledge bases “just in case.” We stripped them down to only what each specific task needed. Her email agent got only the email templates and tagging rules. Her meeting prep agent got only the relevant contact data and agenda format.

Accuracy went back to near 100%. Same agents. Same tasks. Just tighter context.

How to Build an Agent That Actually Performs

When I teach agent building in my workshops, I always start with two questions before anything else: what is this agent's role, and what is its goal?

They're different things, and confusing them is one of the main reasons agents underperform.

The role is like a job title. It defines identity. Tone. How the agent thinks about itself and the work.

The goal is the specific outcome you want. The deliverable.

Think about how you'd hire a human assistant. You'd tell them: “You're a travel coordinator” (role) and “I need you to book me on a flight to New York next Tuesday, under $400, window seat” (goal). Both matter. Skip either one, and the results get sloppy.

Same thing with agents. If you only give a goal without a role, the agent's tone and judgment will be all over the place. If you only give a role without a clear goal, it'll spin in circles.

The formula I use is something I call the OCE approach: Outcome, Context, Expectations. What do you want it to produce? What does it need to know? What format and constraints matter? Once you get these three things defined, most agents start working well on the first try.

What Real Results Look Like

I'll leave you with a story from one of my Two Hour Workday workshops.

Mark was the only student who showed up the day we built a meeting prep agent. (There was a reminder email glitch. Long story.) So he got essentially a private session.

We built him a simple agent that pulled upcoming meetings from his calendar, ran background research on each attendee, and sent him a briefing before every call.

Setup time: about thirty seconds of actual work on his end.

A week later, I followed up.

He told me he'd had several meetings that week and felt genuinely better prepared going into every one of them. He'd even started running it before his church committee meetings. The cost? A few cents a day.

“I've only been on it a week,” he said, “but it's been a lot of fun.”

That's what good context engineering makes possible. Not just better prompts — agents that actually work the first time, cost next to nothing to run, and feel like they were built for exactly your situation.

Because with the right context, they were.

The Quick Version

If you want to start applying this today:

1. Add context before asking — who are you, who's the audience, what format do you want?
2. Keep context tight — only give the AI what it needs for this specific task
3. Define role AND goal — for any agent, both matter
4. Don't blame the model — if output is weak, check your context first

Most AI problems aren't model problems. They're context problems.

And context engineering is way easier to fix than retraining a model.

Want to learn how to build AI agents that actually work for your business? I run hands-on workshops in Austin. Reach out here.

We end up at another friend's apartment. And this person — a crypto developer — shows us something he built. An AI agent that monitors code repositories for bugs, and then automatically negotiates bounties via email. No human in the loop. Just the agent doing the work.

I stood there watching it run.

Monitoring. Deciding. Emailing. All on its own.

And I remember thinking: oh my gosh. This is real.

Not “AI could theoretically do this.” It was doing it. Right in front of me, in this guy's apartment in Ho Chi Minh City.

That was the moment everything shifted.

The Difference Between Using AI and Deploying AI

Up until that point, I'd been a pretty heavy AI user. ChatGPT for drafts. Claude for analysis. Perplexity for research. I was getting a lot of value out of these tools.

But I was still the one doing the work.

Every time I needed something, I'd open a chat window, type a prompt, read the response, copy something out, edit it, paste it somewhere else. The AI was making me faster. But I was still in every loop.

What I saw in Vietnam was categorically different. The agent wasn't waiting for me. It wasn't in a chat window. It was running on its own, watching for conditions, making decisions, and taking action.

That's not a faster search engine. That's a digital teammate.

I came home and spent the next few months rebuilding how I work.

The 63-Cent Podcast Outline

One of the first things I automated was my podcast content workflow.

Before, putting together a podcast outline was a 3-4 hour project. Research the topic. Gather examples. Find the right stories. Write the structure. I'd put it off constantly because it felt heavy.

Now, when I want to start a new episode, I move a card in Jira. That triggers an AI agent that researches the topic, generates a full outline, and pulls from a database of my real personal stories to illustrate each point.

The whole thing runs in about 2 minutes. Total cost: 63 cents.

That's not a typo. 63 cents. For something that used to take me 4 hours.

Why Most AI Content Falls Flat (And How to Fix It)

This is the part I want to spend some time on, because it's where most people get stuck.

If you ask ChatGPT to write a LinkedIn post, it'll write a LinkedIn post. But it'll be generic. Bland. It'll have the kind of “insights” that could apply to anyone.

The reason is simple: the AI doesn't know who you are. It doesn't know your stories, your clients, your specific experiences. So it makes things up. Or it writes something technically accurate but completely impersonal.

I ran into this with a client earlier this year. He wanted to use AI to create LinkedIn content. We set up a workflow, and the posts came out… fine. But they felt fake. Like a press release, not a person.

The fix was building a story bank first.

A story bank is a database of real things you've actually said and done. Stories from client calls. Examples from your workshops. Observations you've shared on podcasts or in meetings. All captured as plain text and stored somewhere searchable.

When my content agents write anything now, they search that database first. So instead of inventing a vague example, they pull something real. A specific client. A specific situation. A specific outcome.

The difference is night and day. The content actually sounds like me.

Start With What You Do Every Single Day

One thing I've learned from building agents for myself and clients is this: don't start with the impressive use case. Start with the frequent one.

I call this the 80-20 approach to agent building. What do you do every day or every week that eats the most time?

For me, it was email. Meetings. Content creation. Those three things alone were taking 6+ hours a day.

For one of my clients — a VC named Evan — it was meeting prep. Every day he'd spend time digging through emails and notes to get context before calls. I built him a daily briefing agent. Now he gets a prep doc every morning automatically. What used to take 20 hours a week of human assistant time now takes minutes.

That's the 80-20 in action. High frequency, high pain. Compounding ROI.

The quarterly-use-case automations can wait. Build for the daily loops first.

What Agent as Teammate Actually Means

The mental model that helped me most is thinking of agents as teammates, not tools.

A tool waits for you. A teammate is aware of what's going on, knows the context, and can act without being told every step.

That Vietnam crypto agent wasn't waiting for its developer to say “hey, go check for bugs.” It was running a loop. Monitoring. Finding opportunities. Taking action within the boundaries it was given.

That's what I aim for when I build agents now. They should know enough context that I don't have to babysit them. They should have clear limits on what they can do without approval. And they should get out of my way for anything they can handle.

The Week That Made It Click

Earlier this year, I tracked what my Lindy agents saved in one week.

239 hours.

That's equivalent to six full-time employees working a full week. In one week. From automations I'd built over the previous few months.

Some of that is email triage. Some is meeting prep. Some is content creation. Some is research. It adds up faster than you'd expect.

I'm not sharing this to brag. I'm sharing it because the number surprised even me. The compounding effect of high-frequency automations hits different once you see it on paper.

Where to Start

If you're reading this and still using AI mostly as a chat tool, here's what I'd suggest:

Pick the thing you do every single day that feels like a drag. Email sorting. Meeting notes. Research briefs. Proposal drafts. Whatever it is.

Start there. Build one agent. Get it working at 70%.

Then let it run for a few weeks and watch what happens to your week.

You don't have to see Vietnam to get the aha moment. But you do have to stop treating AI like a faster search engine.

The agents are the real thing.

Want to learn how to build your first AI agent? Check out the Productivity Academy for hands-on training with Thanh.

“I’d love to use your space. I’m cc’ing my assistant Linda to help coordinate schedules.”

The contact replied. Linda replied back. They went back and forth over a couple of days.

I woke up to a confirmed meeting on my calendar.

Never touched it. Never saw the emails until after the fact.

And when I checked the bill on my Lindy account? Total cost for the entire coordination: 15 cents.

That was the moment this stuff stopped being a cool party trick and started being infrastructure.

Linda Isn’t a Person

When I said I was cc’ing my “assistant Linda,” I wasn’t lying exactly… I just wasn’t being fully transparent. Linda is my Lindy AI agent. She has access to my calendar, can read incoming emails, and knows my scheduling rules.

She saw the original email, checked my calendar, and responded with available times. When the venue contact replied with a different time, Linda checked again and counter-offered. They went back and forth a few times.

Done. Meeting on calendar.

The contact had no idea she wasn’t talking to a human assistant.

Now, if you want to give AI agents actual people’s names, that’s a judgment call. I’ve moved toward being more upfront since then — Hudson Penn (a client of mine who runs four businesses) actually prefers that I mention it’s an agent. He finds it more impressive, not less. But the point is the agent can do this level of coordination without you ever getting involved.

The Numbers Behind My Agent Stack

That 15-cent meeting isn’t a one-off.

Right now I’m running about 45-50 AI agents across my work. On my Lindy dashboard, I can see a weekly report: how many tasks they ran, and how much time they saved.

Last week: 1,074 tasks. 73 hours saved.

The week before: 867 tasks. 54 hours saved.

That’s somewhere between 2 and 3 extra work weeks per week… freed up. Not because I’m working less. Because a bunch of repetitive coordination work that would normally fall to me (or to a human assistant) is just… handled.

The 80-20 Agent Building framework I teach in my workshops is built around this idea: automate the tasks that happen daily and weekly, not the impressive-but-rare ones. Meeting scheduling happens constantly. That’s why it was the first agent I built, and why I still recommend it as the first one for anyone getting started.

The Three Levels of AI Work

Most people I meet are stuck at what I call Level 1.

There are actually three levels of AI work:

Level 1 — AI Assisted. You go back and forth with the AI to produce something. ChatGPT conversations, Claude responses. You’re still doing the driving.

Level 2 — AI Workflows. You’ve built a predictable process. Same input, same output, every time. More reliable, but you still kick it off manually.

Level 3 — AI Agents. The agent runs on its own. It monitors for triggers, executes a workflow, delivers the result. You’re not in the loop at all unless something goes wrong.

Most people I teach are comfortably at Level 1. Some are experimenting with Level 2. Very few have gotten to Level 3 — but that’s where the real hours come back.

The scheduling agent I built for Linda is pure Level 3. It monitors my inbox, finds the emails that need scheduling, handles the coordination, and puts the meeting on my calendar. I don’t initiate it. It just runs.

Your First Agent Should Be This

If you’ve been thinking about building an AI agent but haven’t started yet, start with scheduling.

Here’s why: it’s a contained problem. The inputs are clear (calendar availability, email threads). The output is clear (confirmed meeting). The failure mode is low-stakes (worst case, you step in and sort it out yourself).

The setup I use with Lindy takes less than an afternoon. You define your availability windows, set your meeting preferences, connect your calendar, and give the agent some context about how you like to communicate. After that, you cc it on threads where scheduling needs to happen, and it takes over.

A few weeks in, you’ll stop noticing it. Which is the point.

That’s what I mean when I say AI agents aren’t tools you use — they’re more like teammates you manage. The goal is eventually you don’t have to manage them much either.

The 15-cent meeting was just the beginning. Now my agents handle email triage, weekly transcript synthesis, meeting prep docs, and a bunch of other things I used to spend real time on.

It compounds fast once you start.

Want to build your first scheduling agent? I cover this in my AI workshops and also do one-on-one implementation sessions. You can reach me at thanh@asianefficiency.com or check the Productivity Academy for more resources.

Visit www.asianefficiency.com for more productivity tips and tactics.

DripDrop — Get 20% off your first order at dripdrop.com with promo code tps.

Links

• DripDrop — use promo code tps for 20% off your first order
• 25X Productivity Coaching
• Claude
• Claude for Chrome
• Lindy
• Otter
• Calendly
• OpenClaw
• Manus

If you enjoyed this episode, follow the podcast on Apple Podcasts, Spotify, Stitcher, Overcast, Pocket Casts or your favorite podcast player. It’s easy, you’ll get new episodes automatically, and it also helps the show. You can also leave a review!

The idea: take a podcast transcript, run it through a Lindy agent, and automatically generate a series of formatted slides. What used to take me a couple of hours of manual design work… done in minutes. I was pretty happy with it.

Then I went to update the workflow. New models were out. Gemini 3 and ChatGPT 5.1 had just dropped. Figured it was a good time to upgrade.

Within 20 minutes, I was confused.

The intro section wasn't showing up. The layout I'd dialed in over weeks was completely wrong. I kept tweaking prompts, trying to rephrase things, adjusting the instructions. Nothing worked. Both of the “smarter” models were just… not following my format.

Eventually, I gave up and went back to Gemini 2.5 Flash. The cheapest model available.

First try. Perfect output.

Why Newer Models Sometimes Follow Instructions Worse

This seems counterintuitive at first. If a model is smarter, shouldn't it do a better job?

Here's the thing: “smarter” often means the model has been trained to be more helpful, more creative, more interpretive. It fills in gaps. It tries to anticipate what you probably meant. It improvises.

For a lot of tasks, that's great. You want the model to help you think. To push back. To synthesize.

But for a very specific formatting task — like generating a carousel with a particular structure, every time, exactly right — that interpretive intelligence works against you. The model decides your instructions might be too rigid. It tries to make it better. It changes things.

Gemini Flash was trained differently. It reads the instructions and does them. That's it.

I've started thinking of it like hiring for different roles. If you need someone to process invoices to a very specific format, you don't hire the most creative person in your office. You hire the person who reads the instructions carefully and follows them exactly. Different jobs need different people. And different tasks need different models.

The Automation That's Been Running for 10 Years

Here's a related thing I've been thinking about.

In my Gmail, I have a filter that automatically labels any email containing words like “receipt,” “order,” or “thank you for your purchase” and drops them into a folder called Purchases.

I set that up over 10 years ago.

It still runs every single day. I basically forgot it existed until someone asked me how I keep my inbox clean, and I started listing all the automations I have. That filter showed up on the list and I thought, “oh right… that thing.”

That's what a good automation does. It disappears into the background and just works. You stop thinking about it.

The mistake people make is chasing new tools and new models before the old ones have even finished paying off. The Gmail filter cost me 10 minutes to set up a decade ago. I have no idea how many hours it's saved me since.

A Pattern Worth Noticing

I've been using Lindy for a while now, and I keep noticing that my relationship with it feels similar to something that happened about 15 years ago with OmniFocus.

When Asian Efficiency was just getting started, I built a strong early relationship with the OmniGroup team. They were building something that felt ahead of its time — a serious task manager for serious people. We partnered on content, wrote deep tutorials, and helped their community understand how to actually use it. OmniFocus became one of the most-read topic areas on the site for years.

Lindy feels like that. It's a small, focused company building something real. The product keeps getting better. The team is responsive. And the people who are learning to use it now — really use it, not just dabble — are going to have a significant advantage in the next few years.

The principle here isn't really about Lindy specifically. It's about betting early on things that are working, even if they're not mainstream yet. And sticking with the version that actually works for you, even when newer options arrive.

How to Think About AI Model Selection

Here's a simple framework I've landed on:

When to use a smarter (more expensive) model:

• Research and synthesis
• Writing first drafts
• Analyzing complex data
• Conversations that need nuance

When to use a cheaper, more compliant model:

• Following a specific output format
• Structured data extraction
• Repeatable automation steps
• Anything that needs to work exactly the same every time

The test is simple: does this task need intelligence or compliance? If it needs intelligence, use the best model you have access to. If it needs compliance, the cheaper model probably does it better.

And if an AI tool isn't doing what you want — before you spend an hour rewriting prompts — try swapping the model. You might solve it in 5 minutes.

One More Thing

The automation you set up 10 years ago might still be the most valuable one you have.

Check your Gmail filters. Check your old Zapier or Make workflows. Check the Shortcuts you built on your phone in 2019 and forgot about. Some of those are probably still running quietly, doing exactly what you told them to do.

That's what good automation looks like. You stop thinking about it. It just works.

If you want to dig into AI model selection and workflow design in more depth, that's exactly what we cover in our AI workshops. The next one is May 31 in Fort Worth — details at asianefficiency.com.

People would come to the first one, get the fundamentals, and then… come back again a few months later. I was confused at first. I figured once they had the basics, they'd go figure out the rest on their own.

But then I realized: nobody has time to keep up with AI on their own. The tools change every few weeks. A model that was best-in-class in January might be middle of the pack by March. It's a lot.

And here's the thing — these weren't people who lacked curiosity. They were smart, busy professionals who genuinely wanted to stay current. They just couldn't find 2-3 hours a week to read all the articles and newsletters and watch all the YouTube videos.

So they came back to workshops as a way to stay updated. Which is kind of a compliment, I guess. But it also told me there's a real gap here.

Staying current on AI shouldn't require a workshop every few months.

Here's how I actually do it.

The Walk Trick

Last fall during my AI research masterclass, I showed something that got a bigger reaction than I expected.

I had collected about 15 AI news sources for the week — articles, newsletters, a few YouTube video summaries — and dropped them all into Notebook LM. Then I used the audio overview feature to generate a 20-minute audio summary. Two AI voices, back and forth, covering the key points from all 15 sources.

Put it on my phone. Went for a morning walk. Came back caught up.

That was it. No extra time blocked in my calendar. No reading piled up from the week before. Just a walk I would have taken anyway.

The audio isn't perfect. It misses things sometimes, or summarizes a point too quickly. But 85% caught up on 15 sources beats 0% caught up after skipping another week of reading.

If you haven't tried the audio overview feature in Notebook LM, this is the use case for it. Collect whatever you've been meaning to read — 5 articles, 10 articles, doesn't matter — drop them in, generate the audio, and take it for a walk. The sources can be PDFs, links, YouTube transcripts, text files, anything.

(Side note: I also imported 300 Lex Fridman podcast episodes into Notebook LM once, just to see what would happen. In a few minutes I had access to 600 hours of content I could query by topic. Not something you need to do every week, but it shows how far this thing goes.)

The Daily Briefing I Set and Forgot

The walk trick handles my weekly catch-up. The other thing I set up is a daily briefing.

ChatGPT has a feature called Tasks. Most people don't know it exists. It lets you schedule a prompt to run automatically — daily, weekly, whatever cadence you want.

My prompt runs every day at 11am:

Summarize in bullets the latest AI news. For each item, suggest: (1) how I could create content or courses around it, (2) whether it's interesting from an investing angle, (3) what I should look into deeper.

And then it just… runs. Every day. Without me doing anything.

The briefing shows up in my ChatGPT chat at 11am. I skim it over lunch. Some days there's nothing new. Some days there's something worth digging into. Either way it takes about 3 minutes.

The key is the framing. I didn't ask for “summarize AI news” — I asked for news plus three specific lenses: content creation, investing, and deep dives. That turns a generic news summary into something actually relevant to what I'm building.

You can customize it however you want. Your lenses might be different. The point is making the briefing work for your context, not just a generic download of what happened.

Why I Stopped Using Google for AI Research

This is a little embarrassing to admit, but I honestly can't remember the last time I used Google Search for research.

I asked a room full of people at that October masterclass: if you use ChatGPT or Perplexity, do you still use Google? Almost nobody raised their hand.

That tells you something.

For research — real research, not “what time does this restaurant close” — ChatGPT and Perplexity are just better. They synthesize. They pull from multiple sources. They give you a summary with citations instead of a list of links you have to click through individually.

I use Perplexity most for real-time stuff because it's pulling current information. ChatGPT for synthesis and analysis. Claude for longer reasoning. Gemini for anything involving Google's ecosystem or images.

This is what I call being Multi-Tool Native — not picking one AI and treating it like a hammer, but routing the work to whatever actually fits the job. Different tools have different strengths and you get better results when you use them accordingly.

But the main point is this: for staying current on AI specifically, dedicated AI research tools are faster and more useful than a search engine. The research you do with Perplexity or ChatGPT in 10 minutes used to take 45 minutes on Google.

Putting It Together

Here's the whole system, simplified:

Daily: ChatGPT Tasks briefing at 11am (set it up once, runs forever). Skim it over lunch. Takes 3 minutes.

Weekly: Collect whatever AI content has accumulated (articles, newsletters, YouTube transcripts). Drop into Notebook LM, generate audio, take a walk.

That's it. Two habits. One requires almost no active time. The other turns a walk you'd take anyway into a research session.

You don't need to read everything. You don't need to be expert-level on every new model release. You just need enough signal to know what's worth paying attention to and what you can ignore.

The goal isn't omniscience. It's “good enough to keep moving.”

Want to see how I use these tools in practice? My AI workshops are small-group, half-day sessions where we actually build workflows together — not just theory. Details at asianefficiency.com.

Either they want to copy the whole thing immediately. Or they take one look at the 30-item checklist and say “I could never do that.”

Both are wrong.

Here's what I tell them: I didn't build this in a day. I didn't even build it in a year. The weekly review I run now is the result of 15 years of small additions. And it started with two questions.

“What did I learn this week?”

“What can I do better next week?”

That was it. Every Sunday, same time, same spot. Two questions.

What Actually Happened Over 15 Years

A few months in, answering those two questions started to feel automatic. So I added one thing: a quick look at the week ahead on my calendar. Just to know what was coming.

That became automatic too. So a few months later I added a task review. Then a goals check. Then a project sweep. Each addition happened naturally, once the previous step felt like nothing.

This is how 2 questions became 30 items.

None of it felt like building a system. It felt like a Sunday afternoon routine that kept growing. By the time each new step was added, I wasn't deciding to do it… I was just doing it.

That's what I mean when I say habits on autopilot are a form of automation.

Automation Doesn't Have to Be Technical

Here's where most people get stuck. They hear the word “automation” and picture code. Zapier workflows. AI agents. Something technical and complicated that requires setup.

But the most reliable form of automation is a habit that runs without any thinking at all.

And you're already doing this in other parts of your life.

If you have autopay set up for your rent or utilities, that's automation. You made one decision, set it up once, and now it runs every month without you having to think about it. The money moves, the bill gets paid, you didn't lift a finger.

A consistent weekly routine works the same way. You set it up once (with two questions, ideally). You repeat it at the same time every week. Over time it becomes invisible. It just happens.

The Automation Spectrum

At Asian Efficiency we think about automation in a few different layers. There's life automation — routines, recurring decisions, habits. There's digital automation — templates, filters, shortcuts. And there's AI automation — agents that handle synthesis, research, drafting.

Most people only talk about the third layer. They skip the first one entirely.

But the life automation layer is the foundation. Recurring payments. Morning rituals. Weekly reviews. These run with zero friction once they're established. No app required. No monthly subscription. No debugging when the API changes.

And they compound. Each habit you build makes the next one easier to add.

Why Most People Fail at Weekly Reviews

I get asked about this a lot. People try a weekly review, do it once or twice, then drop it.

The reason is almost always the same: they tried to start with too much.

I used to have the same problem. Back before I had a system, every time I sat down for a review I had to figure out what to do from scratch. What should I check? What order does this go in? That friction was enough to make me avoid it.

The fix was making it a checklist. Open it, follow the steps, done. No thinking required.

But the checklist started with two steps. That's it.

If you've never done a weekly review before, trying to implement a 15-step version on day one is like trying to run a marathon having never jogged before. You'll stop after the first mile and decide you're not a runner.

Two questions. One sitting. Every week.

That's the whole plan for the first six months.

What to Actually Do This Week

Try this tonight or Sunday:

1. Set a 15-minute block at the same time each week (Sunday afternoon works for me, around 3pm)
2. Put two questions in your calendar invite or a note: What did I learn this week? What can I do better next week?
3. Answer both. Write it down somewhere, even just a notes app.
4. That's the whole review.

Don't add anything else for at least 4 weeks. Let those two questions become automatic first.

After a month, if it's feeling effortless, add one thing. Maybe a calendar check for the week ahead. That's all.

In 6 months you'll have a 4-step review that feels like nothing. In a few years you might have 10 steps. In 15 years, maybe 30.

But you'll get there the same way I did. One Sunday at a time.

If you want a more complete guide to building a weekly review that actually sticks, check out the Weekly Review Blueprint — it's the exact system I built over 15 years, documented step by step.

I’d been a heavy Lindy user for over a year. Built dozens of workflows for clients. Taught it in workshops. Pushed it harder than most people I knew. So I sent a message: “I’m in the area. Can I stop by?”

They said yes.

I spent a few hours at their office with the team and the CEO. And by the time I left, I was one of three preferred AI implementation partners in the world.

Three. That’s it.

I didn’t plan that outcome. I just showed up.

Why This Felt Familiar

Back when I was building Asian Efficiency in its early days, I flew up to Seattle to visit OmniGroup — the company that makes OmniFocus.

We started collaborating on workflows. They’d send us beta access. We’d create tutorials and courses around their software. Both companies grew from that relationship. It was one of the best strategic bets I made in the first decade of running the business.

This Lindy visit felt the same way.

The pattern: find a platform you genuinely believe in, go deep on it before everyone else does, and show up in person when most people wouldn’t. That’s how you end up with a relationship that goes past being a customer.

What the Partnership Actually Means

Here’s the gap Lindy has as a company. They build the software. They don’t do custom implementation.

So when a Lindy customer contacts support and asks “can you help me build this workflow?” — Lindy’s answer is no. But now they say: “We know some people.” And they refer to three of us.

That’s it. There’s no formal program. No marketing arrangement. It’s just: when a customer needs hands-on help building AI agents on Lindy, they get referred to me.

The reason I got that spot isn’t because I pitched myself. It’s because I’d already built a real track record. I could show the work. I could talk about implementation challenges they recognized from their own customer support tickets. We were speaking the same language before we even sat down.

What I’ve Been Building With It

One thing I showed them during the visit was a meeting transcript infographic workflow I’d set up with Gemini.

The prompt is almost embarrassingly simple: turn this transcript into an infographic. That’s it.

Within about 10 seconds, Gemini generates a visual summary — decisions made, action items, key themes. I tested it live on the podcast with my co-host Brooks, pulling up a Lindy meeting I’d had that week. His reaction was pretty much what mine was the first time: this is kind of insane.

Think about what that means for webinars. You run a 90-minute workshop, paste the transcript in, and you have marketing visuals in seconds. Your follow-up email looks like a professional content team produced it.

Gemini at $20 a month. Unlimited.

I’ve also been using Lindy to handle things like meeting prep, follow-up emails, and client automation stacks. The platform has gotten significantly better in the past year. The reason I became a preferred partner isn’t just timing — it’s that I’ve been building real systems on it, not just experimenting.

The Bigger Lesson About Platform Bets

Most people wait until a platform has 10 million users before they commit to it. By then, everyone else has already built their expertise and relationships.

The better move is to identify a platform you genuinely believe in, go deeper than almost anyone else, and do it before it’s obvious.

I did this with OmniFocus and Asian Efficiency starting in 2011. I’m doing it again with Lindy and AI automation starting in 2024.

This isn’t about being contrarian. It’s about actually using the tool, building real things with it, and showing up when most people are still just “curious” about it.

I accidentally started an AI agency after teaching workshops for fun. What began as people asking me to implement solutions — instead of just teaching them — turned into a full consulting practice with more work than I can handle. The Lindy partnership is a direct extension of that.

What This Means If You’re a Lindy User

If you’re already on Lindy and you’ve hit a wall with implementation — agents that aren’t working, workflows that need debugging, a vision for automation you can’t quite execute — that’s what I help with.

I work with business owners and operators who want to actually build their AI stack, not just learn about AI in the abstract.

The best way to get started is to reach out directly. Tell me what you’re trying to automate. We’ll figure out if it’s the right fit.

Booking Q2 projects now.

Visit www.asianefficiency.com for more productivity tips and tactics.

Sign up for a $1/month trial period at shopify.com/tps.

Get 20% off your first order: dripdrop.com — use promo code TPS.

Ready to tackle bigger problems? Go to Claude.ai/tps.

Cheat Sheet

• Top 3 Productivity Resources [1:40]
• Why your brain — not your tools — is the real productivity bottleneck [6:55]
• Why spending $40 on tickets might cost you a good night's sleep [8:57]
• The cognitive trap that was wrecking Brooks's sleep before concerts [10:50]
• The billion-dollar version of that thing you do with your gym membership [11:42]
• Why your brain erases all the annoying parts when estimating how long things take [13:27]
• The buffer time trick that turns chaotic schedule blowouts into clean handoffs [21:00]
• Why designing your environment beats relying on willpower every time [23:52]
• What judges reveal about why you make worse decisions after lunch [28:50]
• The counterintuitive reason Brooks banned snacks from his car [31:35]
• The three hidden biases that are skewing your decisions right now [33:19]
• How Thanh uses AI to surface and audit his own past decisions [38:00]

Links

• Shopify
• DripDrop — promo code: TPS for 20% off
• Claude
• 25X Productivity Coaching
• Espresso Display 15.6″
• The Design of Everyday Things — Don Norman
• System Design Interview: An Insider's Guide — Alex Xu
• TPS152: How to Beat Your Brain and Conquer Cognitive Biases
• The Art of Thinking Clearly — Rolf Dobelli

If you enjoyed this episode, follow the podcast on Apple Podcasts, Spotify, Stitcher, Overcast, Pocket Casts or your favorite podcast player. It’s easy, you’ll get new episodes automatically, and it also helps the show. You can also leave a review!

The call ends. Everyone says bye. You close the window. And then you sit there for a few seconds trying to piece together what you just agreed to.

If you're disciplined, you wrote it down somewhere. If you weren't, you're running on memory until someone follows up.

I lived in that gap for years. Then I built something that eliminates it entirely.

The Problem Nobody Talks About

Post-meeting admin is one of the most invisible time drains in knowledge work. You don't schedule time for it. It doesn't show up on a calendar. But it's there — the 5-10 minutes after every call where you have to manually translate what was said into tasks, emails, and reminders.

Multiply that by 5 or 6 meetings a day and you're looking at 30-60 minutes of unscheduled cognitive overhead. And that's assuming you do it right. Most people don't. Commitments slip through. Follow-ups get missed. Someone emails you a week later asking about the thing you said you'd handle.

I was working with a commercial real estate client in late 2025 whose sales team had this exact problem. Their CRM was always behind. Action items from calls weren't getting logged. Follow-up emails were being forgotten. They weren't bad at their jobs — they were just spending 30 minutes after every sales call doing admin work that nobody wanted to do.

We built an agent that changed all of that. When they finish a phone call, the CRM gets updated, tasks are created, and if they promised a follow-up email, it's already drafted and ready to send. The 30 minutes of post-call admin? Gone.

That experience made me want to build the same thing for myself.

Why I Switched From OmniFocus to Todoist

I was an OmniFocus user for over a decade. Not because Todoist is a bad app — it's genuinely good. But I switched for one reason: Todoist has an API integration with Lindy that makes my current workflow possible.

This is something I wish I'd understood earlier about productivity tools. The question isn't “which app is better?” The better question is: “what workflow do I want to build, and which app makes that workflow possible?”

When someone asks me what task manager to recommend, my first follow-up question is always about their workflow. Are you a visual person? Do you like boards or lists? Do you work mostly on mobile? And — especially lately — do you want to be able to connect it to AI automations?

That last question changes the answer entirely.

How the Automation Works

Here's what happens now at the end of every meeting I have.

Within about 2 minutes of hanging up, I open Todoist. My action items are already in there. Not just a summary of the meeting — specifically the things I said I'd do, with due dates set based on what I committed to.

If I said “I'll get this to you by Monday,” there's a Todoist task due Monday.

If I said “let me follow up on that,” there's already a draft email waiting for me in my inbox.

The workflow runs through Lindy. Lindy reads the meeting transcript, uses AI to identify every action item I committed to — not things others committed to, specifically mine — and creates the tasks in Todoist with the correct due dates. If a proposal was promised, it pre-drafts the email based on the meeting context.

I didn't type a single thing.

The setup took me maybe an hour to configure. It's been running for months. It's one of those automations that's so quietly in the background that I almost forget it exists — until I'm in a meeting and I notice I'm not frantically taking notes anymore. Because I don't need to.

The 80/20 Principle for Automation

This is a good example of what I call 80/20 agent building: focus your automation energy on things that happen daily, not things that are impressive but rare.

Post-meeting admin happens every day. Multiple times a day. A one-time ROI is nice, but compounding ROI is where the real leverage is. High-frequency tasks are the best candidates for automation because you capture the return every single day.

The mistake most people make when they start exploring AI automation is chasing the flashy use case. The complex multi-step agent that does something dramatic. But the simple automation that runs 5 times a day saves you more total time than the sophisticated one that runs once a month.

Start with frequency. Build from there.

What You Need to Set This Up

Here's the short version:

1. A Lindy account (they have a free tier to start)
2. A Todoist account (also has a free tier)
3. A meeting transcription tool — I use Fireflies, but Otter works too
4. About an hour to configure the Lindy workflow

The core flow: Fireflies captures the meeting transcript, Lindy processes it, Lindy creates tasks in Todoist. You'll need to give Lindy your Todoist API key and configure the task format you want.

If you're already using Lindy for other things, this workflow should take you 20 minutes to set up, not an hour.

The Bigger Lesson

This whole workflow started because I stopped asking “what's the best task manager” and started asking “what workflow do I want, and what tools enable it.”

That question reframe changes everything. It makes you build backwards from the outcome instead of forward from the feature set. And when you start choosing tools that way, you end up with a system that actually fits how you work — not how the app designer assumed you'd work.

The post-meeting admin gap is small. But small things that happen daily add up fast.

Try it. See what 30 minutes back per day does for you over a month.

I had a lot going on. I'd come into a day with 12, 15 things on my list. Some days I'd knock out 9 of them and still feel behind because of the 6 I didn't touch. That math never made sense to me at the time. It took a while to figure out what was actually broken.

It wasn't my capacity. It was my finish line.

What the Rule of Three Actually Does

Most people hear “pick your top three tasks for the day” and interpret it as a productivity ceiling. A constraint. Something for people who can't handle complexity.

But that's not what it is.

When you commit to three things and actually finish them, you've “won” the day. You can always add more after. Nobody stops you from doing tasks four, five, or six. The rule of three doesn't prevent that.

What it does is give you a clear definition of success before the day starts — so you stop moving the goalposts on yourself.

Type-A people especially struggle with this. High achievers tend to keep raising the bar mid-game. They finish five things and immediately shift focus to the eight they didn't touch. There's no arrival point. Every day ends in some version of “not enough.”

The rule of three fixes that. When you hyper-focus on three things and finish them, two things happen. You move faster because you're not spreading attention across 15 items. And you actually feel done — maybe for the first time in a while.

The Sunday Version of This

I run a version of this same logic every Sunday during my weekly review.

Most people use their Sunday planning to add to the week. What needs to happen? What am I committed to? What am I behind on? And they end up building a week that already feels overpacked before Monday morning arrives.

One thing I added to my weekly review checklist changes that: I ask myself what I can remove from the calendar.

Not what I should add. What should come off.

Most weeks, I find something. A recurring meeting I haven't re-evaluated in months. A call that could be a three-sentence email. A block I put in out of obligation to someone else's urgency. Things that felt necessary when I added them but aren't moving anything forward now.

This single question — “what can I remove?” — is what keeps a week from collapsing under its own weight before it starts.

It's the same logic as the rule of three, just applied at the weekly level. You're not just planning what to do. You're deciding what success looks like and removing everything that gets in the way of it.

Why This Is Hard for High Performers

Here's the thing: if you're someone who genuinely has a lot of capacity, the rule of three feels wrong. It feels like underperforming. You know you can do more.

And you're right. You can do more.

But doing more and finishing things are different problems.

A to-do list with 15 items and 9 completions looks impressive. But a list with 3 items and 3 completions feels different. There's a clarity to it. A sense of control. And over time, consistent completion of the right things beats inconsistent coverage of many things.

I've watched people with half my energy output accomplish twice as much — because they were relentless about what got on their list in the first place.

The filter is the work. Getting the right three things on the list is harder than it looks.

How to Actually Do This

If you want to try it, here's how I'd start:

For daily planning:

1. The night before (or first thing in the morning), identify your three most important tasks for the day.
2. Write them down somewhere visible — not buried in a 20-item list.
3. Don't start anything else until those three are either done or consciously deprioritized.
4. If you finish all three by 2pm, great — you can add more. But you've already won the day.

For your weekly review:

Before you add anything to next week, ask: “What can I remove from my calendar?” Look for meetings you haven't re-evaluated. Blocks you're holding out of habit. Things that were urgent three weeks ago and aren't anymore.

Remove one thing. Just one. Then plan the rest of the week.

Over time, this becomes automatic. You stop letting your calendar be a graveyard for old commitments.

The Finish Line Changes Everything

The most productive people I know aren't doing more than everyone else. They're doing less, more consistently, and finishing it.

That's the shift. Not more capacity. A clearer finish line.

The rule of three isn't a ceiling. It's a definition of success. And once you have that, you can stack as much on top as you want.

If you want a step-by-step system for your weekly review — including the calendar pruning question — check out the resource. It covers how to set up this habit so it actually sticks.

You know the type. Jewelry on someone’s wrist. Necklace draped over a coffee table. That warm, editorial look you see from the brands you actually want to buy from.

Every time the conversation came up, it got pushed back.

Too expensive. Too much coordination. Need to find a photographer, a model, a location. Maybe next quarter.

This went on for months.

What We Actually Did

One afternoon, we just tried something. I pulled up a tool called Nano Banana and uploaded a few of their existing product shots.

Then I described the lifestyle scene I wanted. Bracelet on a marble surface, morning light, simple background. Something that could go on their homepage or Instagram.

Thirty seconds later, we had it.

Cost: 5 cents.

We did it again. Different angle. Different setting. Each one under a minute. Each one looking exactly like the kind of photo they’d been putting off booking.

The rule with AI product photography for jewelry is this: avoid showing faces. Zoom into the product. The lighting, the texture, the detail… it all reads as real because the product is real. You’re just using AI to handle the environment around it.

By the end of that afternoon, they had 40+ new assets ready to upload.

The Shift Isn’t Just About Cost

The obvious win is money. Traditional lifestyle shoots for a small brand can run anywhere from $500 to a few thousand dollars once you account for a photographer, model, location, and editing. At 5 cents per image, the math is almost embarrassing.

But the bigger shift is something else.

For months, “lifestyle photos” sat on a list of things they wanted to do someday. It wasn’t that they didn’t want them. They always wanted them. The friction of organizing a shoot just kept winning.

Remove the friction and the thing gets done.

This is a pattern I see over and over with small business owners and AI. They’re not resistant to improving their business. They have lists full of improvements they want to make. What they don’t have is time, budget, or bandwidth.

AI doesn’t replace the desire. It just removes what was blocking it.

But You Have to Use the Right Tool for the Job

Here’s something I’ve learned from working with a lot of business owners on AI: loyalty to one tool is a mistake.

The best AI users I know route work to the right tool depending on what needs to happen.

For image work, especially product imagery, Gemini is usually the strongest option right now. For writing that sounds more like a human wrote it, Gemini also tends to outperform the others. For anything technically complex, Claude. For daily strategy and research, ChatGPT. For recurring automations, Lindy.

I call this being Multi-Tool Native. Not picking a winner, but knowing which tool to reach for on which job.

The jewelry brand situation was a visual task. So we used a visual-first tool. That made all the difference. If we’d tried to solve it with a text-focused model or a general-purpose chatbot, the results would have been mediocre. And they might have concluded “AI doesn’t work for this.”

The routing matters as much as the tool.

One More Thing That Changed

After we solved the photo problem, I showed them something else.

I used Claude Code to build a simple tool that tracked their product catalog and matched it against their image inventory. So they could see at a glance which products had lifestyle photos and which were still missing them.

Took a few hours. No developer involved. I just described what I wanted and iterated on it.

Before AI coding tools existed, something like that would have meant hiring a developer, writing a brief, going through a revision cycle, and spending probably $2,000 minimum. Now it’s an afternoon project.

This is the thing people miss when they think about AI for business. They imagine replacing their copywriter or their accountant.

But the real ROI is in all the projects you were never going to do because they cost too much or required skills you didn’t have. AI makes those projects possible. The list of “someday” items gets shorter.

Where to Start

If you have a product-based business and you’re not using AI for your visual assets yet, start there. The tools are genuinely good now, especially for faceless shots.

If you’re a service business, think about what administrative projects you’ve been putting off because they’d require a developer or a designer. There’s a decent chance you can build a lightweight version yourself with Claude Code or a similar tool.

The jewelry brand didn’t need me to tell them they wanted better photos. They knew that. What they needed was someone to show them the friction had already been removed.

Yours probably has been too.

Want to learn how to use AI tools like this in your own business? I run one-day AI workshops in Austin and online. Check the Productivity Academy for upcoming sessions.

I had a Lindy notetaker joining my calls. I had transcripts being saved to Google Drive. I had this whole system set up to capture everything.

And I was reading maybe 5% of it.

Somewhere in that other 95% was a comment a client made in passing that I should have followed up on. A recurring blocker that came up three times in a month. A contradiction between what two different advisors told me in separate meetings that I never noticed because I was reading one transcript at a time.

So I built something to fix it.

The Weekly Synthesizer Agent

Every Sunday night, an agent runs through all my meeting transcripts from the past seven days. Sometimes that's 8 meetings. Sometimes 15. It doesn't matter — it reads everything, synthesizes it into a Google Doc, and drops the doc in my inbox before I wake up Monday morning.

Here's what the document includes:

• Executive summary — a paragraph or two overview of what the week was actually about, not just what was scheduled
• Recurring themes — topics that came up across multiple conversations (if something showed up in 4 different meetings, I want to know)
• Key decisions made — things I committed to or agreed on
• Blockers and open questions — things that haven't been resolved
• Books, tools, and resources mentioned — across all calls, so I don't miss the thing someone recommended in passing
• Relationship signals — who I said I'd follow up with, or who seems like they need a check-in

That last category has probably been the most useful for me practically. I miss follow-ups all the time. Now my agent flags them automatically.

The Contradiction Feature

But here's the part I didn't expect to be so valuable.

Contradictions.

Last month, two people told me completely different things about AI-generated content. One person said AI clones and AI-generated content are the future of personal branding. Another person, in a completely different call a few days later, said authenticity is the only thing that builds real trust and that AI-generated content is destroying it.

Neither of them knew the other had said the opposite thing.

But my agent caught both. It flagged them as a contradiction. And suddenly I had a genuinely interesting question to sit with: which one is right? Are both right in different contexts? Is there a third answer?

You can't get that by reading one transcript. You can only see contradictions when something is holding all the conversations at once.

This is what I mean by transcript-first thinking. Your meetings are already producing data — real conversations, real opinions, real signals. The question is whether you're capturing it in a way that lets you use it.

The Context Profile That Makes It Work

There's a second piece to this that took me a while to figure out.

The synthesizer is only as useful as the context you give it. If the agent doesn't know who I am, what my businesses are, or what I care about, it can't tell the difference between a minor comment and something worth flagging.

So I built what I call a context profile.

It's a single Google Doc that I load into every agent I build. It has my name, my values, my business goals, my working style, my communication preferences, my current projects. I used ChatGPT to help me write it. I basically had a conversation with ChatGPT about myself and asked it to summarize everything into a structured document.

The funny part is that ChatGPT wrote a better summary of me than I would have written myself. It asked follow-up questions. It noticed patterns I hadn't spelled out. And when I loaded that document into Lindy, the agents immediately got better. They stopped giving me generic output and started giving me relevant output.

Think of it like the difference between asking a new consultant to help you and asking one who's worked with you for six months. The work is technically the same. But the consultant who knows your business gives you something you can actually use.

This is the Centralized Context principle in practice. Instead of explaining yourself every time you open a new AI session, you build one document that travels with every agent you deploy. One source of truth. Agents stop asking you the same questions. Output gets more personalized. Everything gets faster.

How to Build Your Own

If you want to try this, here's the rough structure:

Step 1: Get your transcripts somewhere readable

You need your meeting transcripts stored somewhere an AI can access them. Google Drive works well. The key is saving them as plain text files — not PDFs, not formatted docs. Just text.

Step 2: Build your context profile

Open ChatGPT and have a conversation about yourself. Tell it your name, your businesses, your goals, your communication style. Then ask it to summarize everything into a structured document. Review it, fill in gaps, save it to Google Drive.

Step 3: Build the synthesizer in Lindy

The trigger is time-based — every Sunday at 9pm, for example. The action is: load the context profile, read all transcripts from the past 7 days, generate a structured summary, write it to a new Google Doc, and share it to your email.

The whole thing took me about two hours to build the first time. And the output on week one was immediately useful.

Why This Matters

There's a concept in productivity called real vs fake work. Reading through individual transcripts one by one looking for insights is real work — it takes time and attention. But having an agent do the first pass so you can focus on the thinking? That's leverage.

The synthesizer doesn't replace your judgment. You still have to decide what the contradictions mean. You still have to choose which follow-ups to actually send. But it surfaces the raw material so you're making decisions with better information.

Your meetings are already happening. The data is already there. The question is just whether you're letting it work for you.

Want to learn how to build AI agents like this? Check out our Productivity Academy for frameworks on AI-powered productivity systems.

He had read the books. Watched the YouTube videos. His theory was simple: if he could just get more hours in the day, everything would click.

We talked for about 20 minutes before I stopped him.

“Let's actually figure out what's broken first.”

Because here's the thing — most people diagnose their productivity problem as a time problem. And then they spend months optimizing the wrong thing.

The Three Currencies

When I start working with a new client, we don't open their calendar right away. We start with what I call the TEA Framework: time, energy, and attention as the three currencies of productivity.

Each one can become the bottleneck. And the fix for each is completely different.

Time is what most people think about. How many hours do you have? Are you overcommitted? Is your schedule chaotic?

Energy is the less-obvious one. It's your physical, mental, and emotional capacity to actually do the work. High energy and you move through tasks like butter. Low energy and even simple decisions feel heavy.

Attention is the scarcest one in knowledge work. It's your ability to stay with one important thing long enough to finish it well. Not just sitting at your desk — actually being present in the work.

The Three Ways This Breaks Down

I've worked with enough people to know the failure states by now.

Overwhelmed usually means time is the real bottleneck. Too many commitments, not enough hours, scheduling chaos. This one is actually rare — it just feels common because it's the most obvious to name.

Exhausted means energy is the problem. The hours are there but your body and mind aren't showing up for them. This is where sleep, recovery, and task-energy matching matter most. I had a year where I made almost no system changes and just focused on sleep quality. It was the most productive year I'd had in a long time. That's an energy fix, not a time fix.

Distracted means attention is the bottleneck. And this is the one I see most often. The work is there. The hours are there. But focus keeps slipping. The environment is set up wrong. The interruption load is too high. The deep work blocks don't exist.

The Benchmark Nobody Talks About

Here's a number that surprises most people.

If you can protect 10 hours of genuinely focused work per week, that's A+ territory. Not 10 hours of being at your desk or having your laptop open. 10 hours of actual deep focus — flow state, making real progress on things that matter.

One hour per day of that kind of work is still a good week. A solid A.

Most people, when they're honest, are getting 2-3 hours of it. Sometimes less.

Nine 10-minute focus bursts don't add up to 90 minutes of real work. That's not how attention works. The depth compounds — but only if you actually get there.

This is why “just wake up earlier” rarely solves anything. You wake up earlier, but if you're sleeping badly, you're just adding more low-quality hours to the front of a tired day.

How to Actually Diagnose It

The question I ask every client is: which currency is running lowest?

Start with energy. Check your sleep consistency for the last week. Notice when you crash during the day — early afternoon slumps usually point here. Run a quick audit: after your main activities, did that give you energy or drain you? The things that drain you are worth examining.

Then check attention. How many true focus blocks did you have this week? If the answer is “I'm not sure” or “maybe one?” — attention is probably the problem. Your environment is fragmenting your thinking before you even notice it.

And finally, check time. If energy and attention seem fine but you're still falling behind, then yes — time management is the lever. Look at over commitment, meeting load, or whether you're doing work you should be delegating.

The Wrong Fix vs. the Right Fix

The reason most productivity improvements don't stick is misdiagnosis.

The guy who needs better sleep downloads a new task manager. The person who needs to protect deep work blocks books more calls because she thinks she needs to “communicate better.” The person who has genuine overcommitment starts a morning routine.

None of those fixes work because they're targeting the wrong currency.

Figure out which of the three is actually limiting you. Then open the right playbook for that problem. And only that one — fixing all three at once is a good way to fix nothing.

Try this: sit down tonight and ask yourself which one felt most limited today. Time, energy, or attention. That's where you start.

If you want a deeper framework for managing all three, the 25X Productivity System walks through the full TEA diagnosis process along with specific interventions for each currency.

What surprised me wasn't the turnout. It was the question that kept coming up after each session: “How do we get more 5-star reviews?”

Every owner I talked to assumed the answer was some kind of front-desk training or a sign at checkout. Most of them had tried it. Most of them had gotten inconsistent results.

Then I shared what Chris Murphy had figured out.

The Stylist Is the One Who Should Be Asking

Chris coaches service businesses. He spent a long time trying to figure out why some salons were racking up reviews while others, doing similar work, were barely getting any.

The answer wasn't the product, the service, or even the timing of the ask.

It was the person asking.

The front desk staff is busy. When a client walks out, they're already processing the next booking, handling the phone, or saying goodbye to someone else. If they remember to ask about reviews, it's usually rushed. And rushed asks don't get results.

The stylist is completely different. They just spent 60 to 90 minutes with that client. They know the client's name, their history, what they wanted that day, and whether they're happy. And they can see it — that moment when someone looks in the mirror and genuinely lights up.

That's the window.

A quick word from the stylist, in that moment, carries ten times the weight of a checkout prompt.

Why This Is Actually a Systems Problem

Once you understand that the stylist is the key person, everything changes.

It's not just a training insight. It's a systems insight. You can build around it.

Some salons do this manually. They brief stylists on how to mention reviews naturally at the end of a service. It works, but it's inconsistent — some stylists remember, some don't, some feel awkward about it.

The ones getting the most consistent results are the ones who've added a light automation layer.

Here's what that looks like:

1. The booking system logs when a service closes
2. A few minutes after checkout, the client gets a short personalized text
3. The text sounds like it's coming from the stylist, not the salon
4. It thanks them for coming in and includes a direct link to leave a review

Nothing spammy. Nothing generic. Just a nudge from the right person at the right time, automatically.

I've seen this structure work across industries — not just salons. Any service business where one specific person does the actual work and builds the relationship. Massage therapists, personal trainers, consultants. The principle is the same.

The Photo Problem That Nobody Talks About

While I was at the Serious Business conference, I ran an exercise that stopped the room.

I asked: “How many of you have photos from client transformations that you can't post — because there's junk in the background, hair on the floor, power cords everywhere?”

Every hand in the room went up.

Hundreds of photos. Thousands, some said. Amazing transformations. Totally unpublishable.

I showed them something. Take one of those photos. Upload it to Gemini. Ask it to clean up the background. Remove the cord. Remove the hair on the floor.

Twenty seconds later: a clean, postable before-and-after.

The room got very quiet. Then a lot of typing started.

That's what I mean when I say the gap between knowing about AI and using AI is the real problem right now. The tools exist. They work. Most people just haven't connected the tool to the problem yet.

The reviews insight and the photo insight are related, actually. Both come from a simple question: where is value being lost right now because of friction?

For reviews, the friction is between the moment of peak happiness and the act of leaving a review. For photos, the friction is between having the shot and having a publishable asset.

AI can remove both.

Where To Start

If you run a service business and you want more reviews, here's the simplest version of this:

1. Identify the person on your team who has the most relationship-rich contact with customers
2. Brief them on asking at the right moment (not at checkout — during the service's natural closing)
3. Set up a simple follow-up text or email triggered by booking completion, written to sound like it comes from that person

You don't need a fancy CRM to start. A basic booking system and a text automation tool is enough.

The bigger idea here — and this is what I come back to in almost every workshop I run — is that most business problems are systems problems in disguise. You don't have a reviews problem. You have a timing and attribution problem. Fix those, and the reviews follow.

Same with photos. You don't have a content shortage. You have a backlog problem. Fix the friction, and suddenly you have assets.

If you want to go deeper on this, check out the Productivity Academy — we cover this kind of systems thinking in the context of AI workflows regularly.

One small fix at a time.

Wait, no. That's wrong. My phone doesn't ring at all.

A Lindy agent does the calling. It dials the clinic. Navigates their automated phone menu. Tells the receptionist that Thanh Pham is on his way and will arrive in about 30 minutes. Then hangs up.

When I walk through the door, my shot is ready.

I save 25 minutes. Every single visit.

Why This Works (and Why It Took Me a While to Build It)

Here's the thing about allergy shots: they have to be freshly prepared. The clinic can't batch them ahead of time. So if you show up without warning, you sit in a waiting room while they mix it.

For a long time, I'd just wait. It was annoying, but what are you going to do? Call ahead every single time? I'd forget. Or I'd be rushing to leave the house.

The first time I thought about automating this, I almost didn't bother. It seemed too small. Too weird. Not the kind of “AI automation” I was supposed to be building.

But that's exactly the point.

The 80/20 of Automation

I've built a lot of AI agents at this point. I have about 40 that I use regularly. And the ones that actually compound in value aren't the impressive-sounding ones.

They're the boring ones.

The allergy clinic call. The weekly briefing from my calendar. The follow-up email that drafts itself. The Todoist task that creates itself from a meeting transcript.

None of these are sexy. None would make a great LinkedIn post about “disruption” or “the future of work.” But they run quietly, every week, without me thinking about them. And they add up.

This is what I call the 80/20 of agent building: automate the things that happen every week, not the things that would look good in a demo.

When I was helping a client named Hudson set up his scheduling automation, the breakthrough wasn't some complex system. It was two constraints he added: he only takes calls between 1 and 5 PM, and he limits himself to three calls per day. Baking those two rules into the agent changed everything. Suddenly the automation worked with his life instead of against it. Small configuration, massive difference.

The best automations respect how you actually live and work. They don't require you to change your behavior.

How I Built the Allergy Clinic Call

The workflow is simpler than you'd think.

The trigger is a calendar event containing the words “allergy clinic.” When that event appears, Lindy fires off a phone call 30 minutes before the scheduled time.

The tricky part was navigating the clinic's automated phone system. When you call them, a robot answers: “For appointments, press 1. For the Westlake location, press 1. For the downtown location, press 2.” And so on.

I had to learn something called DTMF tones. That stands for Dual-Tone Multi-Frequency — it's the technical name for the tones your phone generates when you press buttons. Turns out you can include these in a Lindy prompt to simulate pressing numbers during an automated call.

I called the clinic myself once, timed how long each menu step took, and built the timing into the prompt. “Wait approximately 20 seconds, then press 1. Wait approximately 15 seconds, then press 1 again.” Then I had Claude help me convert my rough notes into a clean, reliable prompt.

The message the agent leaves: “Hi, this is Lindy calling on behalf of Thanh Pham. He's on his way and will arrive in about 30 minutes. Just wanted to give you a heads up. Thank you, have a great day.”

That's it.

The staff at the clinic now recognize the automated calls. When I walk in, they don't ask my name. They just say “your shot is ready” and point me to the chair.

The Part Most People Get Wrong About Building an AI Stack

When I show people my Lindy setup, they see 40+ agents and get overwhelmed. They think they need to build all of that from scratch, all at once, with some master plan.

That's not how it happened.

I started with the default Lindy meeting notetaker template. That's it. Over the next several months, I expanded it one frustration at a time.

My notetaker wasn't joining meetings from my second calendar. So I added a second trigger.

Some hosts don't allow external notetakers. So I connected it to Granola, an app that records meetings locally. When Granola saves a transcript, it auto-uploads to a Google Drive folder, and the same notetaker workflow processes it.

I realized I was manually copying CRM updates after calls. So I added an HTTP request to my CRM.

Each expansion solved exactly one problem. None of them required knowing the next one was coming.

The first thing I'd tell anyone starting with AI automation: don't design the whole system upfront. Use a template. Use it for a few weeks. Let your own frustrations tell you what to add next.

“Life gets better one agent at a time” sounds cheesy, but it's genuinely how this works. The compounding is real. You build one thing, and it reveals the next thing worth building.

Start With the Most Annoying Weekly Task

If you want to know where to start with AI automation, here's the question I use: what's the most annoying thing that happens to you every single week?

Not the most impressive thing. Not the thing that would look good in a case study.

The thing that makes you groan a little every time it shows up.

For me, it was waiting 25 minutes for an allergy shot that was supposed to take 5 minutes.

Now I just walk in and it's ready.

That's the dream. Small, specific, invisible.

If you want help mapping your first (or next) automation, the Productivity Academy is a good place to start. We cover exactly this process — identifying high-frequency pain and turning it into a running workflow.

Or just start with the allergy clinic problem. Whatever annoys you every week. That's your first agent.

On paper, my calendar looked manageable. Padel a few mornings a week, a handful of meetings, some blocks for deep work. Maybe 6-7 hours of “committed” time per day.

But I was constantly running out of day.

So I started tracking what things actually cost me, door to door. Not the time block on the calendar. The whole thing.

My padel session shows 1.5 hours. But I get up earlier, pack my bag, drive over, play for 90 minutes, come home starving, prepare a bigger meal than usual, shower, and then spend 30-40 minutes where my brain is still in sport mode and I genuinely cannot sit down and do focused work. By the time I am actually available for complex thinking, 3 hours have passed.

That is the hidden time tax.

What the Hidden Time Tax Actually Is

Every item on your calendar has two prices.

The first price is what it says on the calendar. A lunch meeting is 1 hour. A coffee is 30 minutes. Padel is 1.5 hours.

The second price is what it actually costs you. Commute both ways. Getting back into whatever you were doing before. The post-lunch fog. The re-read-all-your-emails-to-remember-where-you-were phase. The context switching that quietly follows you from item to item.

Most people only see the first price.

But it is the second price that eats your week.

I started calling this the hidden time tax when I realized that some items on my calendar were showing up at 1 hour but actually spending 2.5 hours of my time when I counted everything. That is a 150% tax.

Why Your Weekly Review Misses This

When you do a weekly review, you typically look at what is on the calendar and ask: does this need to be here?

That is good. But it is not the full question.

The full question is: what does this thing actually take from me?

Because there is a difference between items that are expensive in time and items that are expensive in recovery.

A 30-minute call with the right person might be worth every minute. The same 30-minute call with the wrong agenda might cost you 2 hours of lost momentum because it happened right before you were going to do your best work of the day.

The elimination question I ask in my weekly review is not just “can I remove this?” It is “what is this actually taking from me that is not showing up on the calendar?”

This is part of the TEA Framework we teach at Asian Efficiency. Time, energy, and attention are the three currencies of productivity. Most people only manage time. But energy and attention have costs too, and those costs do not show up in your calendar app.

How to Find Your Hidden Taxes

Here is a practical way to do this.

Pick four or five recurring items on your calendar from last week. Things you do regularly. Not the obvious stuff, but things you assume are fine because they are short.

For each one, track three things:

1. When did you really start preparing for it (not when it began)
2. When did you really finish recovering from it (not when it ended)
3. What was your energy and focus like for the 30-60 minutes after

Then do the math. What is the actual time commitment? What is the energy cost?

When I did this exercise, a few surprises showed up.

Lunch meetings that felt normal were basically writing off my afternoons. Not because of the time… but because of the food and the social energy and the drive home. By 2:30pm I was coasting.

My early-morning padel sessions were great for energy overall, but terrible for deep work that same morning. I would come back from playing and feel physically good but mentally scattered for a couple hours. I eventually moved my focused work to late morning or early afternoon on padel days.

Even certain calls that were “quick check-ins” had a hidden tax because they happened at transitions in my day and reset my mental state mid-project.

The Leading Metric Fix

Once you find your hidden taxes, you have two options. Cut the thing or rearrange around it.

But before you do either, it helps to figure out what activity matters most to everything else in your week. I call this your leading metric.

Here is the idea. If you are a consultant, maybe the one thing that drives everything else is client calls. If you are a writer, maybe it is the hours you spend actually writing. If you are in sales, it is probably prospecting time.

You do not need to track everything on your calendar. You just need to track the one thing that matters most.

For me, a leading metric is writing. The more hours I spend actually writing per week, not responding to messages about writing, not planning to write, but writing… the better everything else in my business goes. So I track that one number and protect it.

When I started auditing my week through the hidden time tax lens, my main goal was: what is eating into my writing time that I have not noticed?

Turns out it was a couple of things I thought were fine but were not. Fixed those. Writing hours went up. Everything else improved.

What to Actually Do

This week, pick one recurring item on your calendar. Something that should be quick or normal. Track it start to finish, including prep and recovery. See what it actually costs you.

You do not need to cut it. But at least you will know the real price.

Most people are making scheduling decisions based on the paper price. Once you see the real price, you make different decisions.

And honestly? A lot of what feels like a time management problem is actually a hidden tax problem.

If you want a structured way to do this kind of calendar audit, check out the weekly review process inside the 25X Productivity System. It walks through exactly this kind of elimination question.

What surprised me wasn't the pattern. It was how rarely I'd been designing my day around it.

I'd been scheduling hard thinking during my energy valleys and easy tasks during my peaks. Completely backwards. And I kept wondering why I felt like I was working hard but not getting anywhere.

That experience set me on a deeper look at energy… specifically, why some days feel overwhelming even when the to-do list is basically the same as a day that felt fine.

Here's what I've figured out.

The Real Source of Overwhelm

Most people treat overwhelm as a workload problem. Too much to do. Not enough time. The solution feels obvious: get a better system, tighten the calendar, prioritize more aggressively.

And sometimes that's right. But a lot of the time, it's not.

What I've noticed — in my own life and with the clients I coach — is that overwhelm is almost always an energy problem in disguise.

Here's the quick version: if you sleep really well for 4 or 5 consecutive nights, it is genuinely hard to feel overwhelmed. The same workload that felt crushing last week feels manageable. You have mental clarity. More patience. Procrastination stops being the all-day fight it usually is.

Flip that around: if you're running on broken sleep and unresolved stress, even a light workload can feel impossible. Your brain is already spending most of its capacity just keeping you upright.

At Asian Efficiency, we think about productivity through three currencies: time, energy, and attention. Most people only ever manage one of them — time. They try to schedule their way out of problems that are actually energy or attention problems. And it rarely works.

Energy Vampires Are Hiding in Your Calendar

Here's the part people don't talk about enough.

Energy vampires aren't just people. They're situations. Patterns. Reactions you keep having without realizing how much they cost you.

The classic version is the person in your life who consistently leaves you feeling flat. Maybe it's someone at work who complains constantly. Maybe it's a family member who always brings up a stressful topic. When you start distancing yourself from those interactions — even a little — you notice the shift quickly.

But it's also things like:

• The conversation thread you've been avoiding for two weeks (the avoidance itself costs energy)
• Getting stuck in traffic every morning because you leave at the same time out of habit
• Starting the day by checking email before you've had a chance to think

I had a client who was convinced he needed a complete system overhaul. His calendar was chaotic, his task list was a mess, and he felt behind on everything.

When we dug in, two things stood out. His sleep had been poor for weeks — he genuinely hadn't connected that to his overwhelm. And every single morning, he was arriving at his desk already frustrated from traffic because he'd never thought to adjust his departure time.

We didn't redesign his system. We fixed his sleep and removed two daily energy drains.

Same workload. Much less overwhelm.

The Subtraction Move

One of the frameworks I use in coaching is the Energy Pyramid — physical, emotional, mental, and spiritual. When someone feels like they're underperforming, I don't start by asking what we can add. I start by asking where the leak is.

Most of the time, it's physical. Sleep, specifically.

And the fix there isn't a new system. It's protecting 4 or 5 consecutive good nights, which almost always requires saying no to something… a late event, a second glass of wine, another hour of scrolling.

Once the physical layer is stable, we look at the emotional layer. That's where the energy vampires usually hide.

The question I like to ask: what's something in your life that consistently costs you energy before you even get to your desk?

Traffic habits. Unresolved conversations. The news cycle in the first 10 minutes of the day. These things seem small. But they compound fast.

What to Try This Week

If you're feeling overwhelmed and you've been looking to your system for the fix, try this instead.

First, count your good sleep nights from the past week. Not just hours — quality. If the number is below 4, that's probably your biggest lever right now.

Second, audit your morning for energy drains. What's the one thing that reliably puts you in a worse mood before 9am? Be honest. Sometimes it's something you've been telling yourself is fine.

Third, subtract before you add. Before looking for a new productivity system, remove one drain.

It's boring advice. I know. But I've seen it work faster than any system upgrade I've ever recommended.

If you want a structured way to diagnose where your energy is leaking, the weekly review process we use at Asian Efficiency walks through this — including how to build a week that actually protects your peak hours.

ChatGPT Pro is $200/month. Claude Max is $200/month. Google Ultra is $250. That adds up. And when you're not sure what you're getting in return, it's a fair thing to wonder about.

But I want to flip the question. Because I think the cost question is temporary. And the question most people aren't asking… is not.

Remember Your “Top Five”?

If you had a mobile phone in the early 2000s, you remember this.

Plans back then charged you per minute for most calls. Texting costs $0.10 to $0.15 each. But carriers started offering “top five” plans — five people you could call unlimited, while everyone else still cost money.

You had to actually think about whether a call was worth it. Is this quick enough? Can I just send a text instead?

Nobody thinks like that now. We call anyone, anywhere, without calculating anything.

I brought this up on the Lightbulb Moments podcast when the host Tom asked about AI costs. Are they going to come down? Is it sustainable to spend this much?

My answer: yes. And faster than people expect. The same thing that happened to mobile minutes will happen to AI compute costs. It always happens with technology. The early adopters pay the “top five” pricing. Everyone else inherits the infrastructure.

But here's the part most people miss.

The Cost Goes Down. The Gap Doesn't.

Back in February, I ran my first AI workshop in Austin. Didn't do any real marketing… just sent a few texts and put something on Instagram. Sold out in four days.

I honestly thought it was a fluke. So I ran another one the following month. Same thing. Sold out in days. Five people came back from the first workshop because they wanted to stay current.

The people in those rooms were not worried about cost. They were worried they were already behind.

And that's the more honest version of what's happening with AI right now. There are two groups: people who are building fluency with these tools, and people who are waiting until it feels more settled, more affordable, more obvious.

The first group is compounding their skills. The second group is building a gap that gets harder to close every month.

AI Adds to You, Not Replaces You

There's a fear underneath the cost question too. A lot of people worry that using AI too much means becoming dependent on it… or worse, outsourcing the parts of themselves that matter.

I actually think about this one differently.

Every December I write handwritten Christmas cards. This is a tradition I do every year, and I take it seriously. I still write every note by hand. But sometimes I'll ask an AI for ideas first.

I'll describe a friend I haven't talked to in a while, and ask for a few ideas of what I could write. Then I'll look at what it suggests and think… “yeah, that one.” And I write my own version in my own words.

The card is still mine. The gesture is still mine. The AI just helped me think of something I might have forgotten to say.

That's how I think about these tools in general. They're a thought partner. A good one. Not a replacement for your judgment or your relationships or your ideas. Just something that helps you think a little better, a little faster.

The AI Fluency Levels

In my workshops, I teach what I call the AI Fluency Levels — three stages most people move through.

The first level is AI Assisted: using chat tools like ChatGPT or Claude to help with tasks. Writing, research, drafts, brainstorming. This is where most people should start.

The second level is AI Workflows: stringing tools together so information flows between them automatically. An email triggers an action. A calendar event creates a prep document.

The third level is Building Agents: creating systems that run on their own, monitor things, and take actions without you triggering them every time.

Most people I meet are ready for level one right now. Some are ready for level two. Very few need to think about level three yet.

The point is not to skip ahead. The point is to start.

One Tweak a Week

The way I actually recommend people start is simple: do one more thing with AI this week than you did last week.

That's it. Not a full workflow overhaul. Not a stack of new subscriptions. Just one thing.

If you've never tried it, start by asking ChatGPT to draft an email you've been putting off. Or ask it to help you prepare for a meeting you have tomorrow. Or describe a problem you're solving and ask it to help you think through the options.

One tweak a week. Let curiosity compound.

The AI costs will come down. The “top five” era won't last. But the gap between people who built skills early and people who waited? That one tends to stick around.

Start now. Even small.