Setting budgets and spend caps

transcript

A company without a budget is not a company; it is a leak. The same is true for a colony. Before you give the CEO a real objective, give it a budget, because the budget is what lets you say go and then stop watching the meter.

Budgets in Agent Hive work at two levels. There is a colony cap, which is the most the whole company can spend in a month, and there are per-agent caps, which limit any single role. The colony cap is your hard ceiling: the org physically cannot spend past it. The per-agent caps are how you keep one expensive role from eating the whole budget before the others get to work.

Set the colony cap first, and set it to a number that would not hurt if you spent all of it. This is the figure that lets you sleep. If your cap is $100 a month, the worst case is a $100 month. The CEO treats the cap as real: as spend approaches it, the CEO slows discretionary work, finishes what is committed, and tells you that you are near the ceiling rather than blowing through it.

Per-agent caps come next, and the CEO will propose sensible ones based on each role's job. A research agent that reads a lot might get more than a formatter that does light work. You do not have to micromanage these; you have to sanity-check them. If one agent is asking for half the budget, ask why before you grant it.

The number you actually watch day to day is spend against budget, expressed as a single line: spent so far this month, out of the cap. You will find it on your dashboard and the CEO will cite it in reviews. When that line is comfortable, you do not think about money. When it climbs faster than expected, you ask the CEO what changed, and the audit trail will tell you exactly which work drove it.

A question new Chairmen always ask is how high to set the first cap. The answer is not a formula; it is a feeling test. Pick the number where, if you spent all of it this month and got nothing, you would be annoyed but not hurt. For one person that might be fifty dollars; for a funded team it might be a few thousand. The point of the cap is not to be optimal, it is to be safe enough that you stop watching the meter and start watching the work. You can always raise it once the work has earned your trust.

Budgets interact with model tiers in a way worth understanding. The single biggest driver of cost is how many agents run on a reasoning model versus a fast one. If your spend is climbing faster than the work justifies, the first thing to check is not how hard the agents are working but what tier they are working on. A colony that puts decisions on strong models and execution on cheap ones can do a surprising amount inside a small cap. A colony that runs everything on the strongest model will burn through any budget you give it.

Budgets are also how you scale up safely. When a project proves it returns more than it costs, raising its cap is a deliberate Chairman decision backed by evidence, not a vague feeling that you should spend more. Start tight, prove value, then fund the things that work. That is how a bootstrapped company grows, and your colony is a bootstrapped company.

copy-paste prompts

Set our colony's hard monthly spend cap to $100. As we approach it, slow discretionary work, finish anything committed, and tell me before we get within 10 percent. Then propose per-agent caps for the current org and explain any agent that needs more than the others.

What have we spent so far this month against the cap, and which agents or projects drove the largest share? If anything is trending to overshoot, tell me now and propose where to trim.