Does a software factory need multiple focused AI agents to run well?
Determine whether specialized roles improve reliability enough to justify their coordination and operational overhead.
Starting point
Gas Town introduced the idea of a “mayor” for a factory: a single point-of-contact agent that knew how to operate it. It also had dispatchers, polecats, a deacon, “dogs,” and other specialized agents. The shape was interesting, but it often produced a great deal of chatter.
Yesod currently uses a Mayor agent and planner agents, and I have experimented with a supervisor agent that intervenes to fix merge issues. The open questions are: What is the optimal set of agents? What roles should they play? Should they be ephemeral or long-lived?
What to track
- Which roles need distinct context, permissions, or lifetimes.
- Which roles can collapse into deterministic services or event-driven jobs.
- How much communication, token use, latency, and operational churn each additional role creates.
- Whether specialization improves delivery and recovery outcomes or merely moves work into handoffs.
- Whether a persistent role performs better than an agent instantiated only when a matching event occurs.
Working hypothesis
A small, stable point of contact combined with focused, mostly ephemeral specialists will retain the benefits of specialization without recreating the chatter of a large, continuously active cast of agents.
Living record
Changes to Yesod’s roster, role boundaries, and operating model belong here.