Tokens per Pull Request

"We just get the bills."

Engineering teams know what Coding Agents cost. They don't know what they're getting out the other end.

AI feels more productive, but is it?

Does the burden of maintaining code the AI wrote eventually outweigh the time saved writing it?

Right now it's all vibes, but as Agentic Engineering continues to become more mainstream the costs will soon need to be backed up with real data.

The enterprises who have rolled out Git AI tried a lot of things first — OTEL, manually tagging PRs, and the heuristics the old guard of developer productivity try to sell them. None of it can tell you what agents actually shipped. To answer that you need accurate AI-Attribution that is baked directly into Git.

Token cost per PR

In the Git AI dashboards you can now drill into any Pull Request and see the total token cost of all the AI sessions that led to it being opened. This includes cloud agents, subagents, and even brainstorming sessions that didn't lead to any code being generated.

Git AI combines sessions from multiple agents into one view — so when Claude Code kicks off subagents, or a developer bounces between Cloud Agent and a local Agent, the full token spend for that PR is in one place.

Tracking AI code after it ships

Getting the cost per PR is only half the question. The other half: was it worth it?

Once AI-written code enters production, Git AI continues to track it so you get a complete picture of how durable that code is, how much rework was required after it shipped, and how many bugs or incidents it leads to.

Especially now — with so much code shipping — you can't just measure PR throughput. You have to measure the throughput of working code. Done has to mean done.

If you want to track all the AI code in your codebase - book a call with the maintainers or get started on GitHub.