Reading the org chart

transcript

Open the org chart and you will see something that looks like a company directory, because it is one. The CEO sits at the top. Under it are function leads, and under them are the specialist agents that do the hands-on work. Each node is a real agent with a role, a model tier, a budget, and a history of work you can inspect.

The first thing to understand is the model tier. Not every agent needs the most capable model. Your CEO and your function leads run on a strong reasoning model because they make decisions. A worker that reformats data or drafts a first pass runs on a faster, cheaper model because it executes. This tiering is why a full org can run for the price of a single seat at most tools. The CEO sets sensible defaults; you can override any agent's tier when you want more horsepower or less spend.

The second thing to understand is scope. Every agent has a job description, and the CEO will not let an agent act outside it without escalating. A support agent does not get to change your pricing. A research agent does not get to send email on your behalf. Scope is a safety property, not bureaucracy: it means a single confused agent cannot wander into a part of the business it has no business touching.

You will be tempted to hire a big org on day one. Resist it. A colony of four agents that each have a clear job outperforms a colony of twenty that overlap. Start small, watch where the work actually piles up, and ask the CEO to hire into the bottleneck. The org should grow in response to real load, the same way a healthy company does.

Trimming matters as much as hiring. If an agent has not done meaningful work in weeks, retire it. Idle agents are not free in attention even when they are cheap in dollars, because they clutter the chart and dilute accountability. The CEO will suggest retirements during reviews; take them.

A useful exercise in your first week is to ask the CEO to walk you through the chart and justify each role. Why does this agent exist, what has it done, and what would break if it did not exist. A role that cannot answer those questions is a role you do not need yet. This is not cruelty; it is the same discipline a careful operator applies to headcount in a real company, where every seat has to earn its salary.

Watch the model tiers as your colony grows, because tiering is where colonies quietly get expensive. The natural drift is upward: it always feels safer to put one more agent on the strongest model. Resist it. The test is simple. If an agent's work on a cheaper tier is consistently good enough, the stronger model is money you are lighting on fire. Reserve the reasoning tiers for the roles that genuinely make judgment calls, and let the executors run lean.

Finally, read the chart as a Chairman, not a manager. You are looking for two things: is there a single clear owner for each outcome you care about, and is anyone overloaded. If an outcome has no owner, that is a gap to fill. If one agent is the owner of everything, that is a single point of failure to split. Everything else is the CEO's problem to manage, not yours.

copy-paste prompts

Walk me through our current org chart. For each agent, tell me its role, its model tier, and the last meaningful thing it did. Flag any outcome I care about that has no clear owner, and any agent that has been idle for more than two weeks.

Where is work piling up right now? If there is a real bottleneck, propose one hire to relieve it, with the role, the model tier, and the monthly cost. If there is no bottleneck, tell me to hold.