Microsoft open-sourced AI Engineer Coach

1/ Microsoft has open-sourced AI Engineer Coach, a VS Code extension that audits how developers use coding assistants rather than generating code itself. The project is published in Microsoft’s GitHub org under the MIT license and is framed as “better agentic engineering.” (github.com) 2/ The extension reads local AI session logs and turns them into dashboards, scores and rule-based feedback. Microsoft says “no data leaves your machine,” and the tool is designed as a read-only analyzer of existing coding-assistant activity. (github.com) 3/ Microsoft says AI Engineer Coach works across “any harness, one dashboard,” with the project specifically describing analysis by workspace, model and harness. The README says it measures usage across coding assistants and can filter views by workspace or harness. (github.com) 4/ The core pitch is discipline. Microsoft says the extension detects 45 anti-patterns across five buckets: prompt quality, session hygiene, code review, tool mastery and context management. Those rules are editable in markdown and can be tested against a user’s own data. (github.com) 5/ That matters because most AI coding tools optimize for speed at the moment of generation. AI Engineer Coach instead tries to score the workflow around the model: how prompts are written, whether sessions are fragmented, whether review habits are weak, and whether context is being managed well. That framing comes directly from the project’s “practice scores” and anti-pattern pages. (github.com) 6/ The dashboard is broader than a simple lint tool. Microsoft says it tracks weekly trends, daily activity, session timelines, generated-code volume by language and model, repeated prompt patterns, and “context health” through readiness checks and instruction-file audits. (github.com) 7/ One notable piece is the rule system. The project includes a Rule Editor and Rule Playground so teams can create, tune and test detections visually or as raw markdown. That suggests Microsoft wants teams to customize the coaching layer to their own engineering norms, not just accept a fixed scoring model. (github.com) 8/ Another piece is reuse. The Skill Finder scans prompt history for repeated requests and flags them as “Custom Skill Opportunities.” Microsoft’s example is a repeated “package the extension” request, which the tool can surface as something worth turning into a reusable instruction file. (github.com) 9/ The project also reaches beyond monitoring into training. Microsoft’s docs describe a Learning Center that generates quizzes and coding challenges from a developer’s actual usage, with progress tracking, difficulty levels and rewards. The stated goal is to turn observed habits into targeted practice. (github.com) 10/ In practical terms, this looks like Microsoft treating AI coding as an operational process that can be measured, not just a chat interface that can be improved with better models. That is an inference from the repository’s emphasis on scores, anti-patterns, audits, timelines and usage-derived exercises. (github.com) 11/ It also fits Microsoft’s broader push to make AI development tooling in VS Code more open. The company previously open-sourced major Copilot Chat components in VS Code, and AI Engineer Coach extends that open-tooling posture into workflow analysis and team coaching. (visualstudiomagazine.com) 12/ The near-term question is adoption. As of the repository snapshot surfaced in search results, the project was public on GitHub with hundreds of stars and recent commits, and Microsoft’s README says installation currently runs through building the extension and installing the generated.vsix package locally. (github.com)