The uphill climb of making diff lines performant

To achieve agent-driven development in Copilot Applied Science, Tyler McGoffin developed a system to automate his intellectual work by creating a coding agent, eval-agents, which analyzes large amounts of code data. He enabled fast development and collaboration by following three core principles: using prompting strategies to guide the agent's thinking, implementing architectural strategies to maintain a clean and well-documented codebase, and adopting iteration strategies that prioritize blameless culture and trust in the agent's outputs. These strategies enabled his team to quickly add 11 new agents, four new skills, and the concept of eval-agent workflows, resulting in significant code changes.

AI automates triage for accessibility feedback, allowing us to focus on fixing barriers—turning a chaotic backlog into continuous, rapid resolutions. The post...