Three projects showed up in my feed within weeks of each other. OpenClaw — 100,000 GitHub stars, a foundation behind it, the creator hired by OpenAI before the dust settled. NanoClaw — runs in a Linux container, connects to WhatsApp and Telegram, built on Anthropic's Agent SDK. PicoClaw — fits on a ten-dollar Raspberry Pi, uses less RAM than a browser tab, 95% AI-generated code.
They all do the same thing: give you a personal AI agent. What changes is the size. And I keep thinking about what that means.
The Pattern
I've seen this before. Software starts big, monolithic, does everything. Then it fragments. Microservices. Unix pipes. The unbundling of every category that got too fat. First you build the thing that does it all, then you break it apart, then the interesting work becomes the wiring between the pieces.
That's what's happening with personal agents. OpenClaw is the monolith — browses the web, reads files, runs commands, does your laundry (almost). NanoClaw is the middleware — isolated, secure, talks to your messaging apps. PicoClaw is the edge node — always on, costs nothing, does one thing.
These aren't competing products. They're the same pattern we've seen a hundred times, just applied to something new.
What Actually Bothers Me
Here's the thing. When you run two or three agents on your machine, each doing different stuff — one researches, one coordinates messages, one runs lightweight tasks in the background — you're not just using tools. You're delegating. You're assigning work to entities with different capabilities and constraints.
That's an org chart. On your laptop.
And it bothers me because I spent years thinking about how organizations delegate, coordinate, and fall apart doing it. Now the same dynamics are showing up at the level of one person and their computer.
If AI is replacing the coordination layer inside companies — and I think it is — then personal agents are creating a new coordination layer for individuals. You're becoming your own middle manager. With all the dysfunction that comes with it.
The Skill No One Is Talking About
Everyone's talking about prompt engineering. Which model is best. Which agent framework to use. But if the future looks like an ecosystem of specialized agents rather than one general-purpose assistant, then the real skill is something else entirely.
It's organizational design. At the scale of one.
Knowing which agent to deploy for which task. Knowing when to orchestrate and when to let them run. Knowing when "good enough on a ten-dollar board" beats "state of the art with full browser access." The same judgment that makes a good CTO or — staying on topic — a good manager.
Which is ironic, because we're automating managers while simultaneously needing to become better ones. Just for our own little fleet of bots.
So What?
I don't know yet. I'm watching people set up these agents and I keep seeing the same mistakes I've seen in organizations for years. Over-delegation. No clear boundaries. Agents doing redundant work. No one checking the output.
We spent decades building organizations to extend what one person can do. Now one person can build an organization on a weekend. Deploy it on hardware that fits in a pocket. Delegate to agents that never sleep.
The question isn't whether personal agents work. They obviously do. The question is whether we'll manage them any better than we've managed the human kind.
I have my doubts.
*Still thinking about this. Probably underestimating something important. As usual.*