Stop Burning Tokens: an AI Coding Agent Runbook

You are not billed once for a conversation. You are billed for the whole conversation on every single turn.

That was the opening line. The session was a practical runbook for cutting token spend in AI coding agents, in three acts:

1. Context hygiene and custom instructions: free, built in and underused.
2. Spec Kit: because rework is the most expensive token leak of all.
3. Lean tooling tricks that keep every turn small.