A VP of Engineering shares a firsthand experiment embedding themselves in a 7-person team for five weeks, using GitHub Copilot exclusively to contribute code. Instead of the planned 2-3 PRs, they shipped 89, covering dependency upgrades, test coverage, CI/CD optimization, and tech modernization. Key takeaways: AI agents dramatically amplify output but human review becomes the bottleneck, agent-generated PR volume will overwhelm teams unprepared for it, and Copilot adoption proved contagious — the team's own PR velocity doubled. The post argues that managers can no longer be passive AI advocates and must become active AI practitioners to credibly drive transformation.
Table of contents
Daniel DoubrovkineSort: