Problem
Our engineers wanted AI coding assistants, but in a security business, source code and customer data need guardrails before any tool touches them. Adoption had to clear Compliance, Legal, IT, and Security, not just Engineering. [TBD: confirm framing; add anything specific that kicked this off.]
Approach
- Worked with Compliance, Legal, IT, Security, and Engineering to bring GitHub Copilot to 50+ developers. [TBD: which company this rollout was at.]
- Led the department’s migration from Copilot to Claude, adding context controls, MCP servers, and secret scanning. [TBD: why we switched.]
- Implemented Vertex AI in GCP, supplemented with Aider, for our internal SOAR and Engineering teams.
- Created a standard for use across the business, with support from GRC and IT.
- Trained teams on managing the context window and getting effective responses.
- Implemented AI PR review in the pull request workflow. [TBD: which tool.]
Outcome
- Cycle time improved by ~14% across the department.
- PR pickup time dropped by an average of 1 day.
- Test writing got faster for standard feature sets.
- [TBD: how cycle time and pickup time were measured, and over what period.]