The chart is a real org
Your colony is structured like a company. The CEO sits at the top, function leads sit under it, and specialist agents do the hands-on work. Each node is a real agent with a role, a model tier, a budget, and a history you can inspect.
Hiring and scoping
You do not hand-build the org. You tell the CEO what you want done and it proposes the smallest team that can do it. Each agent gets a job description that scopes what it is allowed to do. Scope is a safety property: an agent cannot act outside its role without escalating, so a single confused agent cannot wander into a part of the business it has no business touching.
Start small. A colony of four agents with clear jobs outperforms a colony of twenty that overlap. Ask the CEO to hire into a bottleneck only when real work is piling up there.
Retiring agents
Trim what is idle. An agent that has not done meaningful work in weeks should be retired so the chart stays honest and accountability stays sharp. The CEO will suggest retirements during reviews.
Retiring an agent does not erase its history. The work it did, the decisions it made, and the spend it drove all stay in the audit trail, so you can still trace any past outcome back through a role that no longer exists. Retirement is about keeping the live chart legible, not about deleting the record.
Re-tiering an agent
Model tier is not fixed at hire. If a worker on a fast model is producing first drafts that need too much rework, promote it to a reasoning tier and watch whether the quality justifies the cost. If a lead is doing routine work that a cheaper model would handle fine, drop it down. The CEO defaults are sensible, but the chart is yours to tune as you learn where the money is best spent.
A good habit is to re-tier in response to evidence in the audit trail, not on a hunch. If an agent's output is consistently strong on a cheap tier, there is no reason to pay for a stronger one; if it is consistently weak, the cheaper model is a false economy.
One owner per outcome
Read the chart as a Chairman, not a manager. The two questions that matter are whether every outcome you care about has a single clear owner, and whether any one agent is a single point of failure because it owns too much. A gap with no owner is a hire to make; an agent that owns everything is a role to split. Everything between those two is the CEO's job to manage, not yours.
This article is part of the launch docs set; boundaries and depth are still being reviewed with engineering and will keep sharpening.
