Alasdair Allan's QCon London talk, published by InfoQ, pulls together METR trials and an Anthropic study of 52 mostly junior engineers. The Anthropic group using AI scored 17 percentage points lower on a mastery quiz, with the biggest gap in debugging. Allan argues review and supervision are becoming the bottleneck. His suggestions for leads include protecting foundational work for juniors, testing comprehension as well as velocity, and recording architectural decisions along with their reasoning.
