Almost everyone codes with AI now. Almost no one can tell you how much of that code survived.
Git records a human as the author even when an agent wrote the diff. IDE telemetry counts the suggestions you accepted, seconds before you deleted half of them and rewrote the rest. The token bill tells you what you paid, not what you got. So teams run on vibes: the model feels fast, so it must be helping.
We think the honest metric is survival rate: the share of AI-written lines that reach a commit and stay there. An agent writes 1,000 lines, you keep 600, your survival rate is 60%. Run the same task through another model and you might keep 850. That gap is the signal most teams are missing, and it sits before the commit, in churn that no dashboard records today.
What we’re building
Source Trace attributes every line to the operator, agent, and model that wrote it, writes that into git, and turns it into survival-rate analytics. It works with whatever tools you use, whether that’s Copilot, Claude, or a CLI agent, because real developers don’t stay inside one vendor.
Today we’re opening a small private beta. If you run two or three AI tools and want to know which one delivers most quality code, contact us.