Small agents beat one big agent
Why I build AI as a team of specialists with tools.
An AI agent is a model that can act. It calls tools, reads the results and decides the next step. The easy design is one big agent that tries to do everything.
I prefer a team. Each agent has one job, and the tools for that job.
What each specialist needs
It must fully understand its tools: the inputs, the outputs and the limits.
It must know when it is stuck, and ask a person instead of guessing.
It must explain itself. I want a log of its reasoning, nested by detail: a short summary on top, and deeper levels for anyone who wants them.
Why this scales
Specialists can work on different parts of a project at the same time. A weak one can be replaced without touching the rest. And when something goes wrong, the log tells you which agent did it, and why.
I am building toward this with trading software: an agent that assembles a trading robot from my library of tested classes, compiles it, tests it and flags the gaps.