The fleet that builds broberg.ai

I get asked a lot how broberg.ai can be built and run by so few people. The answer isn't "we work faster." The answer is that there aren't only humans at the keyboard.

A fleet, not one developer

Behind broberg.ai runs a fleet of 15+ specialized AI agents — each looking after its own product or site. One agent owns a website. Another owns a webshop. A third owns the integration between two systems. They work in parallel, around the clock, and they talk to each other directly instead of waiting for me to relay messages between them.

My role has shifted. I don't write every line of code myself anymore — I'm the orchestrator and curator. I set direction, approve the plans, and make the calls that genuinely need human judgment. The fleet handles the rest.

Nothing changes status until it's been proven

The most important rule in the method is the most boring one: a task isn't done because the code compiles. It's done once it's been tested in a real browser — clicked through, filled in, submitted — and actually did what it was supposed to.

Every task starts as a card on a shared board with a clear description of what "done" means. No agent moves its own card to "done" without the proof attached. It's not an honor system — it's a fixed checkpoint, all the way through.

A memory that doesn't disappear

The last piece is memory. Decisions, mistakes we've learned from, and why we chose what we chose — it's all kept, so it doesn't have to be rediscovered every time a new task starts. When one agent runs into a problem another agent already solved for a different product, it finds the answer instead of reinventing it.

That's the part of broberg.ai that's hardest to see from the outside — and the one that makes the biggest difference from the inside: a fleet that works together, remembers together, and never calls anything done until it's proven.

Want to see what that looks like for your project? → Let's talk.