You won't find many bigger fans of generative AI than yours truly—and Claude in particular.

Plenty of learned folks believe that single-purpose tools, like note-taking apps, will go the way of the dodo. In short, I agree. I'm on record predicting most niche software vendors will go poof in the near future. But what about enterprise systems? In this post, I'll explain why most of them will survive.

How Are You Using That App?

I will never claim to be an expert on every task management, productivity, and note-taking app out there. Popular choices include Coda, Bear, Obsidian, and Todoist. I do, however, know a thing or six about Notion. For the sake of argument, say that the Notion universe consists of only two groups: basic and power users.

Group #1

A software vendor's app relies heavily upon casual users. The majority of these folks use it in a limited way. Maybe 70 percent of people just scribble down the occasional idea in the note-taking app.

You needn't be a sorcerer to know how this movie ends: Claude Cowork and similar tools will make these simple apps obsolete. (Zapier, Make.com, and other third-party glueware are in similar boats.)

Group #2

Now let's move to power users. I'm talking about people who use the app or system for much more than taking notes. Notion performs dozens of functions for folks in this cohort. Claude, ChatGPT, Gemini, or one of their competitors will ultimately cannibalize such a versatile tool anytime soon.

For Notion power users, the idea that a newfangled AI tool can replace it is absurd.

Now, let's kick it up a level: If a single individual is unlikely to supplant Notion with Claude, then will even a medium-sized enterprise forgo its core systems?

want to learn more about citizen development?

Check out my 2023 book Low-Code/No-Code: Citizen Developers and the Surprising Future of Business Applications.

BUY THE BOOK

A Key Software Distinction

Generative AI tools are inherently non-deterministic. Sure, they'll continue to improve, reduce hallucinations, and ultimately reduce tool sprawl.

By contrast, CRM and ERP systems of record are deterministic. They are far too important to leave to chance. “I usually want to get employees' checks right,” said no payroll manager ever. What's more, these vendors behind these enterprise systems recognize the need for their wares to evolve.

CRM and ERP systems won't disappear.

Connecting the dots leaves one answer: enterprise systems will certainly morph. In particular, their UIs will undergo massive makeovers. It's folly, though, to think that they'll completely disappear. And plenty of others feel the same way.

Claude and Cardio

Apple Health’s data visualizations are surprisingly basic, so I used AI to generate real insights into my workouts.

Phil SimonPhil Simon

Simon's Law of AI Disruption

The more diverse the software, the more likely it is to survive the AI tsunami.

Feedback

What say you?

Perhaps no technology in history has morphed quicker than generative AI. Hell, a week seems like a year for a single company. Even if you spend a good deal of time teaching yourself how to use newfangled tools, it's virtually impossible to keep up.

Against this backdrop, no-code automation giant Zapier recently made news by releasing the second iteration of its AI fluency rubric. TL;DR: If you don't embrace AI, don't bother applying for a job. (Zapier is hardly alone here. Meta grades its workers on their degree of “AI nativeness.”)

Observations

Zapier's post is thoughtful and well-written, but you won't find anything meaningful about how employees can level up their AI proficiency. Noticeably lacking are specifics on any of the following:

Not that long ago, the decision of which application or system to use was typically a no-brainer.

Take the early aughts. Back then, you'd be silly to create a training manual in anything but Microsoft Word. You wouldn't think twice. Ditto for PowerPoint—the de facto standard for presentations. We performed ad hoc data analyses in Excel. You could eschew Salesforce because you preferred your bespoke Microsoft Access CRM. (That choice, though, probably didn't sit well with your boss.) Google's productivity apps worked similarly. You get my drift.

Brass tacks: Only an amateur would use a screwdriver to hammer a nail into the wall.

What about today?

The rise of GenAI has muddied the waters. In today's quick post, I'll offer a few tips on how Claude users can successfully navigate this new world.

Prompt #1

When I begin a new task or project of consequence, I prompt Claude as follows via a Raycast snippet:

When it comes to embracing new technologies, I've always been an early adopter. My friends in grad school still fondly remember my exuberance over the nascent web.

I'll take the Pepsi challenge against anyone. (Thank you, Carnegie Mellon.) At the same time, though, I have to admit: I'm struggling to keep up with Anthropic's crazy pace of innovation.

Recent Additions

As I write these words, here are just some of the features the company has dropped in the last few days:

1. The ability to fully control your computer on whatever device you like—subject to a few caveats. (Cool, but yikes!)
2. Claude Cowork Projects.
3. Claude Code Channels.
4. Memory 2.0.
5. Auto mode to grant or restrict Claude Code's access on your computer. (It's a safer alternative to --dangerously-skip-permissions.

To be fair, you may not have seen any or all of these new releases just yet for whatever reason. Possible culprits include:

• Anthropic's gradual rollout of the feature.
• The responsibilities of your day job.
• A plan restriction.
• A combination of all three.

Maybe it's something else.

Hell, by the time you read these words, Anthropic may have released two more insanely useful capabilities.

Unexpected UI Changes

Yesterday, I was testing Claude's cool new Dispatch feature when I noticed a lightbulb icon in Claude Cowork:

I went all Krusty the Klown:

Even Claude couldn't confidently give me an answer. Here's my prompt:

Reply:

Likely shows. Even Anthropic's own chatbot can't keep up with its torrid pace of innovation.

Looking at the Data

Always curious, I looked at the Claude Code changelog and asked my AI bestie to visualize the number of recent changes in that tool alone. In other words, I omitted Claude Cowork, vanilla Claude, and change Anthropic classified as a fix. Here's the breakdown:

Claude Code Recent Features by Day

Source: Anthropic | For some reason, Claude omitted changes after March 21.

Of course, some enhancements are more meaningful than others. Still, nary a weekday goes by without something new appearing in the Claude suite of tools.

GoDaddy and Microsoft Failed Me. Then Came Claude.

Switching email hosting services neutered my intricate Outlook system. AI helped me find a new provider and reimagine how to prevent spam.

Phil SimonPhil Simon

Simon Says

If I have learned anything over my three decades in enterprise technology, it's this: Tech typically moves much faster than people and organizations can handle. The rise of AI in the last few years reflects that reality in spades.

Is it possible to keep a workforce current on tools that change by the day?

No one is more critical of organizations failing to train their employees properly on new tools than I. (It's a key point in Project Management in the Hybrid Workplace.) At the same time, however, is it possible to keep a workforce current on tools that morph by the day?

In a word, no.

On the diamond, we call individuals who can effectively play more than one position utility players. Fun baseball fact: five people have played all nine in a single game.

I was recently thinking about this term in the context of AI because Claude helped me solve four very different problems before lunch.

Mixologist and Grocery Store Attendant

I know nothing about mixing drinks—even two-ingredient ones. At least I'm self-aware enough to ask Claude:

I learned early on in my career that there is no such thing as perfect software. For decades, I've found different systems and apps limiting in small but critical ways. Out of the box, different tools often can't do exactly what my clients or I need them to do. As a result, I've had to create a variety of workarounds over the years—some of which have lasted far longer than I anticipated.

In today's post, I'll describe one of the most fascinating and powerful aspects of agentic AI: the ability to overcome a software product's core limitations.

The Fleas No Longer Come With the Dog

Notion's native automations are ridiculously useful. I use them with all my clients, but they cannot trigger other automations. Evidently, it would cause a glitch in The Matrix.

Fortunately, there's been a quick, relatively simple fix for a long time:

1. Create an integration with a third-party tool. Zapier and Make.com are popular choices.
2. Link the new integration to the Notion database in question.

Voilà!

Pros and Cons

On one hand, as I write in Low-Code/No-Code, it's never been easier to get these tchotchkes to play nice with other apps. On the other, they add another layer of complexity—not to mention potential technical complications and licensing issues down the road.

Up until recently, most people accepted the limitation and went with the third-party tool. The fleas came with the dog.

The rise of AI agents, however, has changed the game.

For example, consider a simple recurring task. Notion lets you add them to existing databases according to set schedules (daily, hourly, or weekly). If a tree falls in a forest, though, would you hear it? Less abstractly, you need to use one of those third-party tools to alert you of that new task.

Thanks to Notion custom agents, those days are over.

Solving the Problem via AI Agents

Here's a test database after a Notion recurring task auto-populated it:

I had also created a simple notification. If another user added a task to that database, then Notion would notify me:

But here's the rub: That notification would not work on auto-generated recurring tasks without a third-party integration. Without one, I'd never know about the new task unless I manually checked that database. (Again, Notion notifications can't fire based on other notifications.)

To circumvent that limitation without a third-party integration, I created this simple Notion custom agent:

True to form, Notion dutifully alerted me of the new task at 4:15 pm my time:

Total time spent: maybe three minutes.

Technically, a native Notion notification did not trigger the notification. Rather, a Notion custom agent triggered the notification. The semantics don't matter to my clients. More important, I have solved their problem.

Notion Formulas Based on Attachments

Impossible? Think again.

Phil SimonPhil Simon

Simon Says

This is just one of many examples of how AI agents can be useful—and not just in Notion. Stay tuned. They're only becoming more useful. No, they're not perfect, but it's downright silly to ignore them.

In mid-January of this year, I was catching up with a friend on the phone. He asked me if I received his recent email about a job reference. Oddly, I hadn't. Time to do some digging.

I soon learned that I wasn't special. Untold GoDaddy customers were no longer receiving their business emails. The company was experiencing what would turn out to be yet another multi-day Outlook service outage. Even more infuriating, GoDaddy again didn't deign to alert affected individuals. No messages to their backup email addresses. Nary a text. No automated phone calls. Zilch.

GoDaddy's lack of communication irritated just as much as the technical disruption. Just like Pacino in The Insider, I was pissed off and curious. I enlisted Claude to start investigating alternative email providers. Eventually, I decided to kick the tires on Fastmail (affiliate link). As you'll see in this post, I ultimately came out ahead.

An Early Obstacle

I began my migration by changing my MX records. I then exported my emails from GoDaddy to FastMail. Next up, I downloaded my existing blacklist from my personal GoDaddy quarantine page:

To replicate my previous setup, though, I needed to migrate all my existing Outlook email rules. How hard could that be?

Impossible, because Microsoft prohibits Mac users from taking this action. Can someone say vendor lock-in? Maybe the Redmond behemoth only cares about data sovereignty in Switzerland. Ultimately, my hundreds of email filters remain trapped in Outlook and currently unusable.

My MX records propagated. I was now officially on Fastmail, but all new messages arrived in my inbox—not in their carefully curated folders and subfolders. What's more, I received absurd emails from icky domains and scammers that I had previously blocked through my advanced keyword-based filters in Outlook.

I spent about a week intermittently recreating prior filters as I received a variety of messages. By early February, though, I had figured out Fastmail. Why not start noodling with its advanced features?

I soon began to sense that there might be a better way to attack the problem than playing Whac-A-Mole. I then asked myself the following question:

Was replicating my previous rules-based approach to fighting spam really right way to go?

Perhaps my GoDaddy/Microsoft fiasco actually represented an opportunity to build a more sophisticated mousetrap.

A More Modern Approach to Fighting Spam

I'm preparing for my first speaking gig of 2026. The topic shouldn't surprise anyone these days: AI with a dash of citizen development.

As I do with all my talks, I bring my receipts in the form of data. Lots of data. To that end, why not enlist AI to help me generate compelling visualizations?

It's not a difficult ask. Standalone AI image generators have existed since 1973. These days, you need neither a dedicated tool to produce computer-generated JPG and PNG files nor a design program like Canva. Put differently, mainstream AI chatbots are multimodal. As you'll see in this post, they continue to make plenty of baffling non-text mistakes.

Prompting Gemini

I used Claude's deep research feature to generate some recent surveys and studies on AI in the workplace. I threw the results into Notion, took a screenshot of one promising survey, and then crossed over to Gemini. Here's my initial prompt:

Here's its initial response:

That's the data in a structured format. It's a good start, but Gemini failed to produce the requested chart. I responded with the following prompt.

And Gemini then gave me the abomination below:

See a problem?

Sometimes, it's just quicker to pretend that Gemini, Perplexity, and ChatGPT don't exist.

I had to go old school.

AI as a Designer

For a recent post's featured image, I gave Google's much-hyped Nano Banana the following prompt:

Nano Banana required four attempts to return something passable. Pictures of decapitated celebrities aren't my jam.

The Enshittification of Tennis Viewing: An AI Infographic

Explaining the madness that today’s fans suffer when they try to watch a match.

Phil SimonPhil Simon

The AI Paradox Is Alive and Well

So smart one minute and so dumb the next. Expect plenty of hallucinations—and not just in text-based responses.

Such is life in the age of AI. Sometimes, it's just quicker to pretend that Gemini, Perplexity, and ChatGPT don't exist.

Feedback

What say you?

Have you ever talked to your favorite actor, author, musician, celebrity, or athlete? Probably not, but would you fork over a small monthly fee to interact with that individual's digital clone?

Plenty of startups are betting that you will. Here's Claude's quick take on the competitive digital self landscape.

There's no shortage of hype and investment, but can these sophisticated AI chatbots separate fact from fiction?

Let that question serve as the starting point for today's post.

Hello, Arnold

I chose one prominent startup from Claude's list. Delphi.ai allows you to “create your digital self.” You needn't be a celebrity like Arnold Schwarzenegger.

I decided to ask Arnold Bot about the actor's best movie. As you'll see below, the chatbot immediately hallucinated—and not about some obscure Predator or Red Heat quote.

One simple question immediately yielded one incorrect response. At least AI Arnold copped to the error. Perhaps Delphi should train its actor chatbots on accurate and comprehensive actor IMDb pages. Hey, at least Delphi's no outlier. Plenty of other AI tools make basic yet glaring mistakes.

A Second Attempt

I wondered if Arnold Bot would learn from its titanic gaffe (pun intended).

Getting a proper sweat going every day doesn't just keep me in shape; it keeps me sane. I religiously log my workouts via Strava and Apple Health. Those two apps provide basic information on my workouts, not penetrating analysis. (Admittedly, I'm on the free plan for each one, and Apple may enhance its Health app at some point this year.)

Data Deficiencies

At present, neither one lets me easily slice and dice my workout data. Now, I don't need to know the total number of devil's presses I've done. I would, however, like answers to specific questions. Examples include:

1. What's the average number of calories I burn during my Thursday HIIT workouts? Has the number changed over the last few years?
2. What about my walks since last April?
3. I notice the signs at my gym that contain grammatical errors and confusing sentences. Fine, but what else do I do there during a typical week?
4. How much time have I spent getting my ass kicked in Heated Warrior Sculpt since I decided to torture myself?
5. How do my workouts break down by time? A simple histogram would answer that question.
6. How did my 2024 ankle injury affect my exercise routine?
7. Do I really chill on Sundays—my scheduled active recovery?

You get the drift.

Beyond specifics, on a macro level, I can't observe general trends beyond those apps' decidedly limited views.

What to do?

Building a Better Workout Analysis Mousetrap

Fortunately, we live in an era of citizen development. Any curious soul with minimal tech chops can easily create interactive data visualizations, all without a pricey Tableau license or JavaScript background.

After exporting my Apple Health data to a messy XML file on my MacBook Pro, I let Claude—my favorite tool—go to work on the massive file. It created a Python script that parsed the data to a proper, much cleaner CSV format. I ran the script in Claude Code, and it provided summary stats:

want to learn more about citizen development?

Check out my 2023 book Low-Code/No-Code: Citizen Developers and the Surprising Future of Business Applications.

BUY THE BOOK

Interesting, but not remotely visual. Surely we could do better than this simple table. Much better. After all, it's not 1997.

It was time to take it up a notch.

Forty minutes of additional futzing in Claude and Claude Code resulted in a slick interactive dashboard:

A chart or dashboard to answer simple queries is overkill, though. I fired up Claude and asked the dataset questions in natural language.

Of course, you might decide later that you want a graph after all. Ask, and ye shall receive:

Seconds later:

Mouse over a part of the bar chart, and the date and the number of calories burned appear.

Flexes and Limitations

My DIY workout tool represents a vast improvement over Apple Health and Strava, but it's hardly perfect.

For starters, the dashboard doesn't automatically sync workouts. In other words, it won't update after I complete a one-hour GTX class. As a workaround, I would have to either:

1. Run this cumbersome ETL process each time I want to examine my most recent workout data.
2. Pay for the third-party Mac app Health Auto Export or some alternative.
3. Get creative with HealthKit.

In a related vein, my solution also won't alert me if my routine suddenly changes. Finally, it inexplicably fails to count the two or so miles I sprint run on a treadmill during each HIIT workout. Evidently, Apple Health infuriatingly ignores mileage data during these classes. From Claude:

Sorry, Claude, it doesn't make sense. Apple's omission flunks the logic test. (See above.) By my math, the company's decision has effectively cost me over 2,000 miles of treadmill pounding since I started wearing its watch. Call me vain if you like, but I conducted this exercise explicitly to gain better visibility into my workouts.

What a great time to be a citizen developer.

Confidently Wrong

I asked Claude to evaluate my recent Claude Code activities. Its response infuriated me.

Phil SimonPhil Simon

Simon Says

Despite these limitations, my little creation provides insights into my exercise history that Apple Health and Strava—apps with tens of millions of users—currently don't.¹ It's not even close.

I shared the dashboard with a few of my instructor friends. They dug it and want me to show them how to use it after they kicked my ass during one of their classes this week. Along these lines, I decided to publish it on GitHub as a public repo. (My friend and fellow Claude nut Brad Feld gave me the nudge I needed.) Steal it and fork away. Setting it up yourself is pretty straightforward.

If I can vibe code something so useful in a short period, so can you.

What are you waiting for?

Footnotes

In the last few months, I've noodled with AI agents. In a word, the results have been mixed.

To be sure, Claude Code has accomplished some technical tasks that lie beyond my current skills. (I detailed one unabashed success in Monday's post.) At the same time, though, agents have made basic mistakes that I wouldn't expect from a competent tech professional in 2020, never mind today.

A Familiar Script

It turns out that my jagged experience with agentic AI doesn't make me special. Far from it. Asana's State of AI at Work 2025 study reveals that my experience with AI agents is common—maybe even typical.

Here are a few of the report's most interesting stats:

CRMs run the gamut. Behemoths need Salesforce to manage thousands of daily leads. At the other end of the spectrum are the CRM holdouts. Plenty of independent writers and speakers rely on their inboxes to manage a relatively small number of customers and prospects. Hell, even some hidebound small business owners think that email can serve as that type of critical business system. It can't.

I use a modified version of this simple Notion-based CRM. It's not perfect, but it more than meets my needs. If you fill out the contact form on my website, Notion will notify me that you want to discuss a potential speaking or writing engagement or media opportunity.

The Continued Ubiquity of Spam

Regardless of a CRM's size, some inquiries invariably will be bullshit or weak. The fleas come with the dog. Captcha helps, but little prevents determined spammers and scammers from filling out website forms with bum info—or writing scripts to do so.

No surefire way exists to stop all meaningless messages from appearing in your CRM and wasting some of your time. You sure as hell aren't powerless against bad actors, though. Ours is an era of citizen development. One no longer needs to reflexively accept annoyances like spam—at least all the time. Under many circumstances, we can improve our tech. It just takes a little effort, curiosity, and a smidge of stubbornness.

Approach #1: No-Code Automation Tools

I've been doing Notion development long enough to know that I'd have to get creative. In other words, native/vanilla Notion wasn't going to get the job done.

I began my quest to minimize CRM spam by exploring the glorious, magical no-code automation tchotchkes. Without a lick of coding knowledge, anyone can connect different systems and apps in thousands of ways. Zapier, Make.com, and Relay.app (my personal favorite) are just a few popular options.

Zapier, Make.com, and Relay.app can't do everything.

Even when coupled with Notion webhooks, these third-party, drag-and-drop tools can't do everything, though. They suffer from real limitations. For whatever reason, reliably deleting spammy entries from an existing Notion database seems to be one of them.

Approach #2: New NotionAI Features

In theory, Notion's new custom AI agents—now in beta—would do the trick.

In a word, nope.

Noodling with them bore no fruit. I had to manually delete bogus database entries despite Notion's frustrating assurances to the contrary. Different prompts didn't help. After about an hour, I pulled the plug on this method.

Three years ago, I would have called it a day and admitted defeat. Today we all can access an insanely powerful tool.

Approach #3: Agentic AI

Claude Code gives non-developers superpowers. I suspected that, after creating a custom Notion integration, I could use the white-hot agentic Swiss Army knife to:

1. Suggest an appropriate way to automatically purge illegitimate inquiries from my Notion CRM database. (Importantly, Claude Code knows the extent of my technical chops after the last three months.)
2. Help me build it.

Claude Code showed promise right out of the gate. After forumulating a plan with me, it went to work. The agentic AI tool spit out the requisite scripts and even set up the cron jobs for me. I did some testing and the script killed the spam. Forty-five minutes later, we crossed the finish line. Pretty smart.

I decided to release this little doohickey as my first open-source project on GitHub. (Golf applause.)

Confidently Wrong

I asked Claude to evaluate my recent Claude Code activities. Its response infuriated me.

Phil SimonPhil Simon

Go ahead and fork it. You certainly don't have to use it to kill spam in your Notion-based CRM. Perhaps you want to delete leads greater than ten years old. Use it on whatever Notion database and for whatever reason you like. Change the deletion interval. The world is your oyster.

In my previous post, I examined how the role of the analyst has evolved over the past three decades. TL;DR: Much has changed—and the rise of powerful AI tools is perhaps the biggest reason.

Today I'll zoom out and examine some more general questions about AI usage. I'll fuse hard data with some observations.

Is Employee Adoption of AI Tools Increasing?

In short, yes—and rapidly. Late last year, Gallup reported that nearly half of US employees had dipped their toes in the pool. That's nearly twice as many as two years prior.

My Take

No shocker here. We've known for more than six decades about the technology adoption life cycle. From PCs to the web to smartphones to Big Data, there are always early adopters and laggards.

How Are Employees Using AI Tools?

Here are some interesting stats from a February 2025 Pew Research study of more than 5,273 employed adults using AI chatbots:

• 57% use these tools for research and information-finding.
• 52% for editing written content.
• 47% for drafting written content.

My Take

A year later, I suspect that employees are using AI to do far more than act as de facto copy editors and ad-free search engines. Agentic AI has arrived in earnest. Think of tools that can independently take action. And don't get me started on model context protocols. No, MCPs haven't been around all that long, but they are letting non-techies do things that would have previously required years of programming.

What’s Inhibiting Further Employee AI Adoption?

In some cases, it's the threat of being fired. From a Cisco January 2024 survey, 27 percent of respondents claimed that their employers at the time prohibited them from using GenAI applications.

Agentic AI has arrived in earnest.

Another factor, undoubtedly, however, is employee training—or, rather, the lack thereof. Boston Consulting Group in June 2025 discovered that nearly one-fifth of employees received no training on how to use these newfangled tools. Perhaps this fact explains why AI adoption among frontline workers had stalled. It essentially remained constant at 52 percent from 2024 to 2025.

My Take

Again, I'm shocked that the training stat is not even higher.

To be fair, a dearth of employee training does not explain everything related to uneven AI adoption. I've seen this movie before. I can recall many times during my career instances in which management refused to hire proper trainers to teach employees how to use new tech. Senior execs expected employees to pick up new apps and skills on their own time. Paradoxically, these folks would gripe that “the workforce wasn't embracing new tools.”

Are Employees Following Their Organizations’ AI Policies?

The short answer is sometimes. A better question, though, is whether workers are even aware of the rules they should be following. Gallup found that nearly one-quarter of employees don't know what—if anything—their employers are doing with AI.

My Take

I'm surprised that the percentage of oblivious employees who don't know isn't even higher. I have never seen a general-purpose tech arrive with such fanfare and change so rapidly. No doubt that many internal comms and HR folks struggle to keep current on such rapid developments.

Claude’s Third Report Card

Did it bounce back from its first two substandard performances?

Phil SimonPhil Simon

Simon Says

Has generative AI made inroads in many organizations? Absolutely. In a way, though, its rise mirrors prior technological trends: loads of hype, uneven adoption, plenty of opportunity, and no shortage of unmitigated disasters.

For decades, organizations have hired business or data analysts. I should know. In the mid-1990s, I entered the corporate world and held several of them before spreading my wings.

It’s fair to say that the role has permeated organizations for decades—a trend that will continue for the foreseeable future. In 2024, the Bureau of Labor Statistics reported that more than one million people held the role of business analyst in the US. What’s more, the department expects robust growth over the next decade compared to other roles. A quick Indeed search reveals no shortage of analyst roles.

So, nothing has changed in the past 30 years, right?

Hardly.

In today’s post, I’ll describe how the analyst role has morphed over the past three decades.

Yesterday’s Analyst: A Trip Down Memory Lane

It’s fascinating to look at the history of tech and questions. Most germane to this post:

How did firms frequently staff for the analyst position back then?

Here’s an archived job description for a business analyst in 1998. Give it a read. As expected, you’ll see plenty of 20th-century artifacts:

It turns out that Y2K concerns were largely overblown. At the time, though, it caused many executives a restless night.

The Job: All About Data

Analysts’ specific titles varied by employer, but one responsibility remained fairly constant: they worked with data. Lots and lots of structured data. Analysts’ managers heavily leaned on them to:

1. Gather it—typically from IT departments and legacy systems.
2. Cleanse it. (In my experience, the larger and more mature the organization, the older the systems, the worse the data quality.)
3. Analyze it.
4. Prepare static reports, charts, and recommendations on it. These reports typically explained what happened—not what is likely to happen.

It’s essential to note that analysts needed their IT departments to make enterprise data available. Calls to support desks were common. Ditto for the the IT-business divide. (Again, I was there.) Also, few people talked about analytics back then. Finally, business analysts by and large were not statisticians trained on advanced prediction methods.

Automation and Tools

As for automation, ambitious analysts could reduce manual work by creating macros in different Microsoft applications. Visual Basic proved invaluable to folks willing to get their hands dirty—if their employers didn’t ban it.

Against this backdrop, it should be no surprise that analysts spent a great deal of time in front of their ugly beige boxes and now-primitive laptops.

When it came to requirements, most analyst positions listed the typical tech stalwarts of the time. If you didn’t list proficiency in Microsoft Excel and Access, recruiters probably weren’t calling you back. Knowing Structured Query Language proved enormously useful. Less common applications included Crystal Reports and the early business-intelligence tools from Cognos: Impromptu and PowerPlay.

For the most part, the arrows in analysts’ quivers remained relatively stagnant. Notable exceptions included:

• The advent of pivot tables in Excel 5.0 circa 1993. (Yes, they blew my mind back then.)
• The introduction of the infamous ribbon in Office 2007 that confused millions of users.

Brass tacks: the analysts of yesterday spent a great deal of time doing manual work. Sure, their responsibilities were important, but few people would call it strategic.

What about now? Has the position changed? If so, then how?

Today’s Analyst

To be sure, Microsoft Excel has evolved, and it remains many analysts’ go-to app. Don’t take my word for it, though. In 2025, Alteryx released a fascinating survey titled The 2025 State of Data Analysts in the Age of AI. The company gathered data from 1,400 analysts across the globe. Among its most relevant findings here:

• 76 percent of analysts surveyed still rely on spreadsheets for data preparation tasks.
• Over the course of a typical work week, they still spend an average of 10-11 hours collecting data from multiple sources and preparing it..

Of course, the reasons vary. Based on my consulting experience, data quality remains a challenge in many firms. Forty-six percent of data analysts in the Alteryx study reported that it represented the greatest challenge when preparing data.

So nothing has changed, right?

Hardly.

Vastly Improved Tools

For starters, the tools available to analysts today are an order of magnitude better than they used to be. (Whether their employers choose to adopt them, however, is another matter.)

Their benefits of embracing them are manifold. Who wouldn’t want a vast reduction in the time it takes to prepare data for analysis? Interactive data visualizations can obviate dozens of static reports, email attachments, and, ultimately, poor business decisions. Beyond better data prep and visualization, many other powerful, data-related applications exist today that were pipe dreams back then. The reasons vary, but the explosion of artificial intelligence and automation sits at the top of the list.

Contemporary tools can make sense of unstructured and semi-structured data far better than 90s-era tools can.

Even better, today’s business applications are more accessible than ever. As I write in my book Low-Code/No-Code: Citizen Developers and the Surprising Future of Business Applications, one needn’t possess mad coding chops to make them hum.

An Example: Easier Data Prep

As but one example, Alteryx Auto Insights allows analysts—or anyone else, for that matter—to quickly glean valuable insights from boatloads of data. Here’s an example:

Simon Says: Today’s Analysts Are More Strategic

Put everything together, and the role of the analyst has evolved a great deal over my career. Sure, they’ll still need to wrangle and fetch data, but not nearly to the same extent as they did a quarter-century ago.

As a result, they can be far more strategic. Alteryx found that roughly 90 percent of data analysts surveyed agree that their role impacts strategic decision-making and that their influence on business decisions has increased in the past year.

At some point over the past year, you may have heard of the buzzy Anthropic product Claude Code and understandably thought, “This isn't for me.” After all, its name implies that only seasoned or aspiring software developers should fire it up. In fact, nothing could be further from the truth. Plenty of non-techies have discovered that they can use it to do a bevy of cool things—even if they've never heard of an if-then statement or a while loop.

Momentum behind Claude Code has been building for a while. At some point last month, though, it entered the zeitgeist. Articles in The Atlantic tend to signify cultural relevance.

I'd already gone all in on Claude. Why not play around with its powerful AI Swiss Army Knife? After going down the YouTube rabbit hole in early January, I got the hang of Claude Code. I soon started using it to:

• Create bespoke apps and websites.
• Make stylistic changes to this website. (You may notice that I retired the bright orange accent color in favor of a more subdued hue.)
• Analyze data.
• Build new custom Claude skills and tweak existing ones.
• Create and modify local Microsoft 365 files on my MacBook Pro.

That's not to say that I batted 1.000, though. Along the way, I made plenty of mistakes and wasted thousands of tokens in the process.

I was curious about how I could use Claude Code more efficiently in the future. That straightforward request proved far more eventful than I expected.

Asking Claude for Honest Feedback

In this vein, I prompted Claude on January 31, 2026, as follows:

Long story short: It tore me a new one. When it came to Claude Code, I was all hat, no cattle. I suffered from commitment issues. (That is, I never followed through on my promise to use it.) I couldn't follow basic instructions when installing software.

About halfway through reading Claude's jarring and lengthy critique, I started thinking about this recurring question in The Big Lebowski.