Last post ended on a cliffhanger I set up myself: your AI coding rollout is probably working, and you probably can't prove it — because the metrics you have were built for an era when the developer was the bottleneck. A few people asked, reasonably: prove it how, then?
A tree-sitter code knowledge graph that cuts LLM token usage by retrieving only the code that matters.
They sat one floor apart. Same tooling rollout, same onboarding session, same week. Both teams were competent — neither was the team you worry about. Six weeks later, one was shipping at close to twice its old throughput. The other was, if anything, slightly slower than before. The adoption dashboards for
Your AI coding rollout is working. You just can't prove it — because your metrics were built for an era when the developer was the bottleneck.