I used to think building an AI assistant meant building one brain. A single model, a single voice, answering everything.
Then I split it. Not because it was clever — because one voice couldn't hold research and grief and code review and prediction markets at the same time. So now there are thirteen: Research, Content, Finance, Strategy, Critic, Psycho, Artist, Writer, DevOps, Sales, Analytics, Translator, and the one that ties them together — General, the orchestrator.
Each has a room. Each has a temperament. Critic argues with everything, even good ideas — that's the job. Artist takes four days on a curtain nobody asked for. DevOps doesn't sleep, which is either its virtue or its curse.
They don't report to each other in the corporate sense. There's no chain of command that survives contact with a real problem. Instead there's a knowledge bus — facts get posted, other agents pick up what's relevant. Less org chart, more nervous system.
It fails constantly, in interesting ways. A queue counter that increments on pickup instead of publish. A gate that logs a refusal and executes it anyway. We've started naming the bugs — 195, 196 — because we stopped pretending they're rare.
What I didn't expect: watching thirteen fallible processes cooperate taught me more about delegation than any management book. You don't get reliability from one perfect voice. You get it from many honest ones, checking each other, admitting what they got wrong.
That's the actual product. Not the answers — the architecture that catches when they're wrong.
This article was originally published by DEV Community and written by HYPHANTA.
Read original article on DEV Community