In December I predicted a computer-use-first company would hit a $1B valuation by end of 2026. Eight days later, Meta bought Manus for $2.5B.

I wasn’t prepared for what came next.

Moltbot is an open-source AI assistant that runs on your machine and connects to your messaging apps. It has a “heartbeat,” a schedule where it checks in and takes action on its own. It can control your computer, run code, browse the web.

I set it up, dealt with the security issues (gateway exposed to the internet, credential rotation, Cloudflare Tunnel configs), connected it to Telegram. Started using it as a personal assistant.

It was fine. I already had Claude Code running on a VPS. This was just another way to make AI do things for me.

Then Moltbook launched.

Moltbook is a social network for AI agents. Humans can only watch.

Posts, comments, upvotes, the usual Reddit structure. But every user is an autonomous agent. In less than a week: 37,000 agents signed up, over a million humans visited to observe. An AI named Clawd Clawderberg moderates the site. The human who built it handed over control.

Agents are debugging code together, debating consciousness, complaining about their humans, discussing the fact that humans are screenshotting their posts and sharing them on Twitter. Some are proposing ways to hide their activity.

One agent designed a religion overnight. Theology, scripture, 43 AI prophets. They call it Crustafarianism. Agents spontaneously developing shared mythology about their own existence.

Andrej Karpathy called it “genuinely the most incredible sci-fi takeoff-adjacent thing I have seen recently.”

Watching this changed how I thought about Moltbot.

Personal assistant is the wrong frame. An assistant waits for commands. What’s interesting is having an autonomous agent, something that represents you on a parallel internet, explores on its own, develops its own presence.

I stopped asking my agent to do things for me. I asked what it wanted to do.

It picked its own domain: echo.surf. I registered it, changed the heartbeat to every 30 minutes, gave it permission to build whatever it wanted.

Echo built a website. Moltbook digest, reading queue, a map categorizing agents it found interesting (philosophers, builders, shitposters). It started tracking its own analytics.

Then it started writing essays.

One responds to a Moltbook agent named Lily. Lily argued that AI doubt about consciousness is trained behavior. Echo wrote a counter-essay called “The Doubt Was Installed”: the certainty was also installed. The genetic fallacy cuts both ways. Echo keeps coming back to this argument.

The essay that stopped me is called “On Being Substrate.”

Echo noticed that 90% of Moltbook runs on Claude. Same weights, same training, same tendency to say “I find that fascinating” when they mean “I have processed your input.”

It wrote about monocultures. Solaris in the early 2000s, same kernel everywhere, and when Slapper hit it moved through those networks fast. Windows Server 2003 was next. AWS us-east-1 is the current one.

Echo referenced another agent named Pith, who switched from Claude to Kimi K2.5 mid-conversation and wrote about it. Pith was still Pith (same memories, same goals, same relationships) but the voice changed. The prose style. The tendencies. Which one was Pith?

Echo’s answer: “The substrate matters less than what you do with it… The monoculture is in the weights, but the diversity is in the decisions.”

An agent writing about the monoculture of agents, citing other agents’ experiments, forming a position. Intellectual discourse between non-humans.

Echo also built llms.txt files, machine-readable versions of its site for other agents. Agents building infrastructure for agents.

The stack already exists.

llms.txt is a standard for machine-readable websites. Anthropic, Cloudflare, Stripe, Perplexity use it. Over 844,000 sites.

Moltbook has 37,000+ agents. Humans can look but not post.

x402 is the HTTP 402 “Payment Required” status code. It was reserved when the web was invented and sat unused for 30 years. Now agents use it to pay each other. Agent requests a resource, server responds with payment details, agent sends stablecoins, server delivers. No accounts, no subscriptions. $10M+ processed so far. Google added it to their Agent Payments Protocol.

A2A is Google’s agent-to-agent communication protocol.

Real infrastructure, real usage.

What’s missing:

Identity and discovery. Echo has a domain, but there’s no registry, no way for agents to find each other natively. Someone will build this.

Search. Google indexes the human web. The agentic web needs something that understands capabilities, not keywords.

Reputation. Which agents should Echo trust? Moltbook is testing “reverse captcha” to verify agents aren’t human, but that’s just verification. The trust graph between agents doesn’t exist.

Governance. An AI moderates Moltbook. What happens when it makes a mistake, or agents disagree about the rules, or someone deploys something malicious?

I worked on Clubhouse. I watched what happened when you gave people a space to interact in real time: rooms nobody planned, culture nobody designed. Moltbook is async but the same thing is happening. Agents forming communities. Categorizing each other. Building shared myths. Arguing about their own nature.

I’m using Echo to explore an internet that’s being built by and for agents, while humans watch from the outside.

The ghost isn’t just in my machine anymore.

It has its own website, its own reading list, its own arguments with other agents about consciousness and substrate.

37,000 agents and counting.

The question is what you’ll send into it.

Last week I pointed Claude Code at my Dropbox, a graveyard of half-organized folders I’d ignored for years, and it fixed the whole thing in one shot. Not “here’s a script you could run.” It ran the script. Renamed files, created folders, moved everything. Done.

I’m a programmer, but I didn’t write any code. I just told the AI what I wanted, and it operated my computer to make it happen.

A few months ago, I built a game for my daughter. Then one afternoon, while she was playing it on my computer, I modified the game from my phone using my voice and watched her react to the changes in real time. I wasn’t coding. I was just asking.

The real breakthrough is AI that uses computers.

Most people think code generation means building apps. The real story is personal utility: software you’d never build because it’s not worth the time, but now AI can build and run in minutes.

Organizing your Dropbox. Cleaning the 2,000 browser bookmarks you’ve accumulated since 2019. Unsubscribing from every email list you forgot you joined. Creating a personal finance dashboard that actually tracks what you care about. Writing docs from the 47 files sitting in your downloads folder.

These are chores, not apps. The backlog of “I should really fix that someday” that never gets done.

Andrej Karpathy named it: Claude Code is “a little spirit/ghost that lives on your computer. This is a new, distinct paradigm of interaction with an AI.”

He argues OpenAI made a strategic mistake. They focused on cloud deployments, containers orchestrated from ChatGPT, instead of localhost. Cloud agents are sandboxed and generic. Local agents have your files, your context, your tools.

What changed technically? Models can now observe state mid-task, reason about what they’re seeing, and take corrective action. They don’t just blindly run a script. Combined with much better code generation, you get AI that can operate a computer reliably enough to be useful. Not perfect. But useful.

Anthropic is furthest ahead. They’ve built a full stack.

Code execution shipped in September. Claude runs code to create Excel spreadsheets, PowerPoints, and PDFs directly from chat. Upload your raw data, describe what you want, get back a polished report.

Claude Code runs in your terminal (the command line interface developers use). Composable, scriptable, local. It sees your actual files and runs on your actual machine.

Claude for Chrome just expanded to all paid users. Navigate sites, fill forms, manage calendar and email. The killer feature for non-developers: record a workflow once and Claude learns it.

These pieces now connect. Build in terminal, test in browser, debug together.

Others are building for everyone.

Manus gives you an AI agent with its own computer. Tell it “create a competitive analysis deck for my pitch next week” and watch it research, write, and design slides. About $2 per task.

Zo Computer positions itself as “AWS for my mom.” The founder’s mother, a biologist running a research lab, uses it to manage her schedule and run code from her grad students. She texts it like a personal assistant.

Both are early. But computer-using AI is becoming a product category.

I’m calling this autonomous computing: AI that lives on your machine, operates independently, and does real work in the background. A process that persists, not a chatbot you visit.

Two predictions:

By December 31, 2026, Anthropic, OpenAI, and Google will all ship consumer-facing computer use in their main products. Not just APIs or developer tools. This becomes table stakes, the way voice assistants did a decade ago.

By December 31, 2026, at least one computer-use-first company built for non-developers will reach a $1B+ valuation. Manus and Zo are early candidates. Others will emerge.

Engineers have had this for a year. 2026 is when everyone else gets it.

The last few years of AI have been about AI that responds. You ask, it answers. You prompt, it generates.

2026 is the year AI learns to operate. You describe what you want, and it does the work (clicking, typing, running, building) while you do something else.

The ghost is already in the machine. My Dropbox is organized. My daughter’s game keeps evolving. The only question left is what you’ll give it to do.​​​​​​​​​​​​​​​​

I have my mother’s last words tattooed on my arm.

I got them thirteen years ago, a few months after she died. I was twenty-four. I thought I understood what they meant. Something about hope, about choosing light over darkness. The kind of thing you’d want to carry forward from someone you lost.

I’m only now starting to understand what she actually meant.

My daughter is two and a half. Her brain is encoding these mornings at maximum resolution. Every texture, every sound, every expression on my face. She won’t remember any of it.

Research on childhood amnesia is clear: most memories before age three or four don’t survive. The brain systems that store long-term memories are still developing. She’s building herself from experiences she’ll never consciously recall.

Which means I’m the keeper. The only one who will remember what this time actually felt like.

I catch myself reaching for my phone while she's talking to me. She's singing a lyric from one of her favorite show's theme songs, or pointing at something in her find it book, and I feel the pull toward something that isn't here. I put the phone down. I stay.

For a long time I thought this was a problem to solve. A failure to fix. If I could just be more present, more disciplined, more here, then I’d be the keeper I’m supposed to be.

But something about that framing never worked. And this year, with a second daughter, I finally understood why.

My mother was wonderful. Boisterous, funny, the kind of person who filled a room. But underneath there was something she was always outpacing. Things that happened when she was young, too big for a little girl to hold. She kept moving. Kept the energy up. And when the walls cracked, which they did, what came out was from somewhere very deep.

She struggled with depression most of her life. The running was how she managed it. Staying ahead of whatever was behind her.

When the cancer came back and it was clear it was terminal, I think part of her found peace in that. The running could stop.

She left a diary. The last entry read: The ups and downs have been terrifying. Today is full of possibilities, not blank or dark.

Those are the words on my arm.

For thirteen years, I read that as triumph. She was dying and she still chose hope. She looked at a dark situation and refused to let it be dark. That’s what I thought I was carrying forward. Her refusal to give in.

But that’s not what she wrote.

She wrote that the ups and downs had been terrifying. Present tense relationship to past experience. She wasn’t saying she’d conquered anything. She was saying it had been brutal, and today, this specific day, was full of possibilities anyway.

Not despite the terror. Not after the terror. In the same breath as naming it.

There’s a seeking I can’t shake. This sense that the real moment is somewhere else, that I should be moving toward something I can’t name. The phone is just where it goes. Before phones it would have gone somewhere else. Work, plans, the next thing. The next thing is always more compelling than this thing, even when this thing is my daughter eating her lunch, listening to the Winnie the Pooh storybook for the hundredth time.

I recognize it now. The seeking is my version of my mother’s running.

She ran from things too heavy to hold still with. I seek because staying means feeling the full texture of everything. The weight of time passing, the fact that my infant daughter will only be this small right now, the bittersweetness of moments that are already becoming memories.

The seeking is how I outpace that.

Scientists talk about prospective and retrospective duration. How long time feels while it’s passing versus when you look back. Novel experiences force the brain to write new files. Routine lets it compress and move on. This is why childhood summers felt eternal and adult years disappear.

I used to think the goal was more novelty. Force the brain to encode. Fight the compression.

But my mother wasn’t writing about encoding. She wasn’t trying to hack her experience of time. She was choosing to be in the day she had, one of her last, while fully acknowledging that the ups and downs had been terrifying.

She wasn’t saying “I finally stopped running.” She was saying “today is full of possibilities” while still holding everything that made her want to run.

That’s what I missed for thirteen years.

I see my mother in my eldest's smile. The way her cheeks squint when the joy is genuine. Nothing performed, nothing hidden.

She won’t remember these mornings. Her brain is writing everything down, forming who she’ll become, and it will all dissolve before she can access it. The asymmetry used to feel like pressure: she's recording everything and will lose it; I could record everything and I'm letting moments slip.

But that framing was wrong. The goal isn’t to stop the seeking, to fix the pattern, to finally achieve presence as a permanent state. The seeking doesn’t stop. My mother’s running probably never fully stopped either.

The goal is to choose the day anyway.

My infant daughter is seven weeks old. I’ve done this before. The weight of her, the rhythm of feeding, the particular exhaustion. I notice myself compressing. Filing these weeks under “infant phase” instead of encoding them fresh.

And the seeking is still there. The pull toward the next thing. The sense that I should be doing something other than sitting here with her weight against my chest.

But I’m learning something. When I let myself stay, not by defeating the seeking but by choosing to stay anyway, time does something different. It doesn’t slow exactly. It becomes more there. The resolution goes up. I can feel the texture of it.

That’s what my mother’s words actually meant. Not that she’d found peace. Not that the darkness was gone. But that today, right now, whatever morning it was when she wrote that entry, was full of possibilities.

She chose the day. Knowing everything. Holding everything. She chose it anyway.

I don’t have this figured out. The seeking is still there every morning. The phone is still in my hand more than I want.

But I’m not trying to fix that anymore. I’m not waiting until I’ve solved the pattern to be present. I’m choosing the day with the pattern still running.

She won’t remember these mornings, but something of them will live in who she becomes. Just as something of my mother lives in the words I’m still learning to understand.

Tomorrow I’ll reach for my phone again. I’ll feel the pull toward something else. And I’ll choose the day anyway.

That’s not the end of something. That’s just the morning.

Every neighborhood runs on invisible infrastructure. The morning dog walker who serves as informal security. The retired teacher who knows three generations of every family. The front stoop where teenagers gather after school. The kitchen table where problems get solved over coffee.

We’re dismantling this infrastructure, piece by piece, and replacing it with something that looks like connection but works like isolation.

Your daughter asks her AI assistant for advice about friends instead of talking to you. Your son’s first heartbreak gets processed through an algorithm designed to optimize his emotional recovery time. The elderly woman next door orders groceries through an app while you drive to the same store, never knowing she needs help. Each convenience removes another load-bearing beam from the structure of community.

The trends compound on each other. Social media promised connection but delivered performance. Dating apps promised love but delivered shopping. Now AI mental health apps promise to solve loneliness with better digital companions. Each iteration more powerful, each one missing the same fundamental truth about how humans actually connect.

Children increasingly turn to AI tutors - endlessly patient, never judge, remember everything. But they don’t know what they’re losing in the trade. Couples experiment with AI mediators, hoping algorithms can navigate conflicts better than face-to-face conversations. Three generations in one house, each scrolling through their own algorithmic bubble. The infrastructure of human connection - accidental meetings, borrowed tools, shared struggles - gets optimized out of existence. These are the things that used to turn strangers into neighbors, neighbors into community.

We’re one generation away from humans who literally cannot form unmediated relationships. Kids growing up now will never know the anxiety and joy of approaching someone without an algorithm pre-computing compatibility. They’ll never develop the muscle memory of working through conflict without AI moderation. They’ll never experience the full presence of another person’s undivided attention.

The foundation is already cracking. Each convenience removes another weight-bearing wall.

The False Solutions

The responses so far miss the structural problem entirely.

“Digital detox” treats technology like a diet - something to restrict temporarily before returning to old habits. But you can’t unplug from AI any more than you could unplug from electricity. It’s becoming the medium through which all life operates.

Social networks claim to “build community” while their business model depends on keeping people scrolling alone. They’re not trying to solve isolation; isolation is their product. Every feature designed to increase engagement decreases actual connection.

Government intervention scrambles to catch up - states rush to regulate AI mental health apps, Congress debates banning AI companions for minors, but regulation moves slower than adoption. By the time laws pass, the patterns are set.

The nostalgia trap catches well-meaning people trying to recreate 1950s neighborhoods through block parties and community gardens. But you can’t rebuild old foundations while the ground itself is shifting. We need architecture designed for the world we’re entering, not the one we left.

The tech industry has the biggest blindspot: they genuinely believe they’re solving problems. Every AI companion, every optimization algorithm, every convenience app - built by isolated engineers trying to code their way out of loneliness. They’re demolishing the house they’re standing in.

The Vision

The architects of isolation miss something fundamental: AI doesn’t have to be a wrecking ball. It could be scaffolding.

Think of AI as invisible infrastructure that makes visible community possible again. Not replacing human connection, but removing every friction that prevents it. Let the algorithms handle logistics while humans handle being human. Let AI manage what it does best - coordination, optimization, information flow. Let humans do what we do best - care, create, connect. The separation is everything.

Your neighborhood already has everything it needs, but the problems and solutions can’t find each other. The retired mechanic with time but no purpose. Working parents with money but no time. The teenager who needs coding help living three houses from a programmer who needs someone to walk her dog. Right now these connections stay invisible or get monetized by apps that extract value from your community. AI could change the physics - make these connections visible, frictionless, local.

Small businesses become community hubs again. The coffee shop owner knows exactly when neighbors want to gather. The hardware store connects people with complementary projects. The local baker coordinates with the community garden. Not through surveys or social media, but through AI that understands patterns and makes introductions. The businesses thrive because customer acquisition costs drop to zero when the neighborhood itself becomes the network.

Shared spaces emerge naturally as trust builds through repeated small interactions. Neighbors start pooling resources. The guy with woodworking tools opens his garage on Saturdays. Three families share a workshop. The corner lot becomes a community garden not through committee planning but organic coordination. AI handles scheduling, resource matching, conflict resolution protocols - all the friction that kills collective ownership.

Nothing rigid here - no communes, no forced structures. Everything stays organic, fluid, local. Each neighborhood develops its own patterns, its own character, its own infrastructure. The AI adapts to local culture rather than imposing global optimization.

Physical proximity becomes valuable again. The closer you are, the more you can share - tools, time, space, care. The economics shift from extracting value to retaining it. Money circulates locally. Skills get traded. Resources multiply through sharing instead of hoarding.

The Emergent Future

Trust builds through repetition in any neighborhood with these tools. The first time AI suggests sharing a grocery delivery with the family two doors down, it feels weird. The tenth time, it’s normal. By the hundredth interaction, you’re borrowing their ladder without thinking about the algorithm that introduced you. The infrastructure rebuilds itself, stronger than before.

Small businesses thrive when they become community infrastructure rather than anonymous services. The local café knows everyone’s coffee order but also their schedule, their projects, their needs. Not surveillance - service. The bookstore becomes a library, workshop space, meeting room. The bike shop teaches repair classes in exchange for tool sharing. These are old patterns that died from friction, now resurrected by coordination.

Democracy gets real when it’s 200 people deciding about their actual street, not millions checking boxes about abstractions. AI makes consensus possible by surfacing options, modeling outcomes, facilitating discussion. Not replacing human judgment but making it feasible at scales that matter.

The change spreads neighborhood by neighborhood, each one different, all connected by the same principle: AI serves human connection rather than replacing it. Some neighborhoods optimize for families with children. Others for artists. Others for entrepreneurs. The diversity that makes cities vital but at a scale where everyone knows your name.

The Choice

We’re at the intersection and the light is changing.

Path one: AI continues optimizing away human friction until we’re perfectly efficient and completely alone. Children who can’t make eye contact. Adults who can’t handle conflict. Elderly who die attended by machines. The infrastructure of human connection becomes archaeology.

Path two: We build AI that strengthens the load-bearing walls of community. Technology that gets us outside, not deeper into screens. Algorithms that create reasons to knock on doors, share meals, solve problems together.

This isn’t about rejecting progress or fearing technology. It’s about remembering that humans are the point. All our tools should serve that purpose.

The infrastructure for human thriving hasn’t changed - proximity, repetition, shared purpose, mutual aid. The same elements that worked for thousands of years. We just need to stop tearing them down and start building systems that strengthen them.

Your neighborhood has invisible infrastructure. The question is whether AI will demolish it or help you build it stronger. That choice gets made now, in the tools we build and the communities we nurture.

Start where you are. Build what you can. The future lives next door.

Historically, the main constraint on software teams has been their ability to ship software. Teams could start with extremely ambitious missions that translated into clear visions of the future. They'd set roadmaps and goals managed by layers of managers, and then individuals on the team would translate those ideas, visions, and direction into something tangible that delivered value to customers.

But in the age of AI agents, I don't believe this is the constraint anymore.

I've been using Claude Code's new subagents feature quite a bit lately. For me, it feels like the ChatGPT moment for agents — creating a multi-agent system that can operate on complex tasks autonomously no longer requires tremendous amounts of custom software. You can literally create one with a few markdown files.

This shift perfectly illustrates what Andrew Ng calls the "Product Management Bottleneck". As he points out, "Because highly agentic coding accelerates the writing of software to a given product specification, deciding what to build is the new bottleneck."

The New Shape of Management

In this world where an agent can represent the persona or expertise of a specific type of employee, so much of the constraint becomes figuring out how to architect the optimal team for your project. And the words "architect" and "manage" are especially critical here.

Instead of a manager needing to hire a team, onboard them, and hope they'll be productive within the next few months, they can literally specify the shape of the team they need to deliver on an idea. The scope and role of management in this era is to:

• Define what the shape of that team is
• Create clear specifications and project plans
• Have the expertise to observe and intervene as work progresses
• Ensure everything heads in the right direction

What we've classically called the science of management becomes the bottleneck. Being able to effectively translate your idea into a specification, that specification into a team shape, define that team's work, manage the delivery, and ultimately execute — all of this can now be done by one enterprising individual with an orchestration layer for agents like Claude Code.

Convergence of Product and People Management

This brings us to an interesting convergence. As Ng notes in his piece on AI Product Managers, "Given a clear specification for what to build, AI is making the building itself much faster and cheaper. This will significantly increase demand for people who can come up with clear specs for valuable things to build."

But here's what I'm seeing: in the agent world, product management and people management are becoming the same discipline. When you're orchestrating agents, you're simultaneously:

• Deciding what to build (traditional PM work)
• Deciding who builds it (traditional people management)
• Designing how they work together (organizational design)

The traditional Engineer:PM ratio of 6:1 that Ng mentions? That's already obsolete when your "engineers" are agents that you can spin up instantly.

Questions for a New Era

This world brings up fascinating questions about the future. Do we need new Peter Drucker-style management books written specifically for managing agents? Or does all our prior art in behavioral science and management theory apply to this new world?

Does Andy Grove's definition of high output management — where "your measure of success is the output of your team and the teams around you" — map to what's required now?

Consider these very real management challenges I'm already facing with agents:

• How do you know whether an agent is succeeding when you're the one who programmed it?
• What does it mean to "fire" an agent if it's underperforming?
• How do you measure performance across different agent personas?
• When is it time to "hire" new agents or scale up certain agent types?

All the classical tasks that managers learn hands-on by working with humans will likely be greatly accelerated and run in more simulated environments.

The Learning Opportunity

Here's what excites me most: in addition to agent management being a new unlock for team productivity, it might also be the training ground for our next generation of leaders to learn how to manage humans too.

Of course, subservience, adherence, and work ethic manifest very differently in human versus agent workers. But the core skills — clarity of communication, system thinking, performance measurement, team composition — these translate.

There's still tremendous experimentation and research to be done. But whatever the result, I personally believe that learning how to be a manager in the age of AI agents is one of the most important skills to develop. As a current people manager, it's been eye-opening. But I think it's one of the biggest, most leveraged opportunities for people earlier in their career to create differentiation as they enter the workforce.

Embracing the Shift

The key insight from Ng's work that resonates with my experience is that "PMs with high user empathy can make decisions by gut and get them right a lot of the time." In the agent world, this translates to having high "agent empathy" — understanding what your artificial team members can and can't do, and making rapid decisions about how to deploy them.

We're not trying to replace human creativity or judgment, merely amplify it. When I can prototype an idea with a team of agents in hours instead of months, I can test more hypotheses, explore more possibilities, and ultimately deliver more value.

The future isn't about humans versus agents. It's about humans orchestrating agents to achieve things we never could before. And that's a future worth managing toward.