Simon Green argues that cheap AI output leads teams to over-build code, tests, policy and process. Creating an artefact is now almost free. Understanding, maintaining and retiring it is not. He draws on Steve Yegge's account of Wheelhouse, an agent system built to develop the game Wyvern. Wheelhouse runs dozens of agents and has grown its own constitution, rulings and enforcement checks. Yegge added an officer role purely to prune obsolete rules. Green calls this capacity seeking utilisation: idle agents get fed work, which creates coordination failures, which create more machinery. He cites Faros telemetry showing more tasks and pull requests but longer reviews and more bugs. A BetterUp survey found many desk workers receive plausible-looking AI work that shifts effort onto them. His remedy is ordinary discipline: budgets for internal machinery, reasons for every rule, retirement paths, and measures taken outside the factory.

• 600,000 lines in the agent system
• 270 commits a day
• 450 governance artefacts
• 20-25% of work spent on the factory