
Tools the teams built themselves - and today most teams work with them.
The problem: Strong AI infrastructure, a few advanced teams - and big gaps across the rest.
What we did: My method - a workshop on the team's own code, close hands-on work, and then the team leads. Each team builds its own agents and skills, with an AI Champion and one metric chosen by leadership.
- Skills the teams built themselves now run in development and in the on-call rotation - saving time and catching problems early.
- A developer built a tool that learned from six months of comments by the strictest reviewers. The tool fixes the code before it reaches review.
- Manual alert triage became a documented process, and was the team's starting point.
- A pilot of 2 teams, today more than 10 - leadership expanded the program team after team.

