AI Coding Agent Cost Ledger: Find Expensive Sessions Before They Become Normal

Short summary

A practical guide to building a cost ledger for AI coding agent sessions that goes beyond simple token tracking. It explains how to capture model requests, tool calls, retries, approval waits, and value evidence per session, then connect spend to outcomes like merged PRs. The goal is decision-grade visibility so teams can spot wasteful patterns—such as rereading unchanged context—before they become defaults.

• •Coding agent sessions hide cost in retries, repeated context reads, and tool-call loops
• •A cost ledger tracks tokens, tool calls, wait times, and value evidence per session
• •Connecting spend to outcomes enables better model routing, context limits, and pricing decisions

Generated with AI, which can make mistakes.