Your AI coding agent gets expensive one reasonable decision at a time

Short summary

AI agents become expensive through thousands of small, individually defensible decisions—reading extra files, using powerful models for simple lookups—that feel free in isolation. Context bloat and poor model selection are silent cost multipliers that compound on every turn and don't reset. The author built a discipline structure using hooks, routing, and delegation that unexpectedly improved both cost and quality.

• •Agent costs accumulate invisibly from small decisions that feel individually reasonable and harmless
• •Context bloat compounds silently—bloated sessions re-bill their costs on every subsequent turn
• •Structural discipline (hooks, routing, delegation) reduces costs while improving agent correctness and focus

Generated with AI, which can make mistakes.