In an ancient Greek household, the oikonomos was the steward: the one trusted to manage the estate’s resources so the whole household prospered. The word is the root of economy, which began as the art of running a household well.
Every engineer on your team has their own agent setup right now. Their own system prompts, tuned over weeks. Their own skills, written for their own habits. Their own idea of what a good spec looks like once it hits the agent.
Nobody asked them to build this. It happened the way local tooling always happens - one engineer solves a problem for themselves, and the solution stays theirs.
The difference is cost. A better linter config doesn’t show up on an invoice. A better system prompt burns fewer tokens on every run, for the rest of that engineer’s time on the team. Multiply that across a team, and the gap between the best-tuned loop and the worst one is real money, invisible on any dashboard anyone is looking at.
We got disciplined about cloud spend years ago. Tagging, budgets, alerts, someone whose job is to notice when a service starts costing more than it should. Token spend is heading the same direction. Almost nobody has built the equivalent muscle yet.
