How I run my side projects with a fleet of AI agents

I have more side projects than hours. So instead of doing everything myself, I run them with a small fleet of AI agents. This post is the first thing I’m writing about it, and this site was built the same way.

What is the fleet?

It’s a handful of agents with clear jobs:

• A chief of staff. One agent keeps the big picture: what matters this week, what’s waiting on me, and who should pick up what.
• Specialists per project. Each project gets its own designer and developer agents. The designer proposes options and mocks; the developer turns the one I pick into a pull request.
• Routines. Some work just needs to happen on a schedule: checks, reminders, small recurring chores. Those run on their own.
• A shared lessons doc. When something goes wrong, the fix gets written down in one place every agent reads, so the same mistake doesn’t happen twice.

What do I still do myself?

The decisions. Agents propose, I pick. Nothing ships until I say “ship it”, and anything that goes out under my name gets my eyes first. My job looks less like typing code and more like editing: choosing between options, saying what feels wrong, and deciding what’s next.

Why work this way?

Because it lets me ship. Ideas that would have sat in a notes app now get a first version, then a second. And the habits that make human teams work well (clear owners, small changes, writing things down) turn out to matter just as much for agents.

I’ll write more here as I go: what works, what breaks, and what I’d do differently.