How it works
Each agent is a small program with one job, its own memory, and its own written instructions — instructions I can edit and version without touching code. It can pick up tools, hand work to a helper, and decide for itself when it's done. The lines between them are strict contracts: the Creative Head can only hand the Director a brief in exactly the shape the Director expects, so nothing gets lost in translation. The ring around everything is a switch — a person can put any agent on manual, or stop it mid-run, without a deploy.
Why a team
One giant prompt that plans, makes, checks and remembers turns into a monster nobody can debug. Five small agents, each doing one thing, can be read, tested and fixed one at a time — and when one starts misbehaving you pause just that one while the rest keep working. The ring isn't decoration; it's the reason I can sleep.