Cost per authored PR is a period's estimated coding agent spend divided by the pull requests your engineers authored in connected repositories over the same period. VibeCheck, Maestro AI's coding agent visibility product, calculates it for a team, a person or the whole organization, across Claude Code, Codex and Cursor Desktop Agent sessions, once the GitHub App is connected.
The Problem
Total spend only ever goes up as adoption grows, so on its own it reads as bad news. Larkspire Health, a fictional 25-engineer team, doubled its agent spend in a quarter. The finance partner's question was fair: did the work get more expensive, or did the team just do more of it? Nobody could answer from the invoice.
How It Cascades
Rising spend with no denominator looks like waste, even when output rose with it.
Teams can't compare this quarter with the last on equal terms.
Leaders reach for anecdotes about a few great sessions instead of a number finance recognizes.
The Insight
Spend needs a denominator people already trust. Authored pull requests are counted in GitHub, not estimated by a model, which makes them a fair, boring yardstick to put next to agent cost.
The Solution
Connect the GitHub App. VibeCheck counts the pull requests each engineer authored in connected repositories and shows cost per authored PR beside total spend in Usage, and on each Person and Team page. Switch the view between tokens and estimated cost, and pick 4, 8 or 12 weeks or the weeks you want. It's deliberately simple: total period spend divided by authored PRs. It doesn't try to split a session's cost across pull requests. Where there's no PR data for a period, VibeCheck says so and shows a dash instead of a zero.
The Outcome
Larkspire's lead brings two numbers to the quarterly review: total agent spend and cost per authored PR for each quarter. The second one is what tells finance whether the spend is getting healthier.