microsoft/AI-Engineering-Coach: better agentic engineering

better agentic engineering

Summary

AI Engineer Coach is an open-source VS Code extension (also runnable as a canvas in the GitHub Copilot app) that reads local AI session logs and surfaces insights about how you use coding assistants. It is licensed under MIT, written in TypeScript, and requires VS Code 1.115 or newer. All parsing and analytics happen on the local machine, with no telemetry sent out.

What it is useful for

The tool helps engineers who regularly work with AI coding assistants understand and improve their usage patterns. It breaks down AI-generated code volume by language, model, workspace, and harness; charts practice scores with weekly trends and daily activity; and flags 45 anti-patterns across prompt quality, session hygiene, code review, tool mastery, and context management. It also scores context health via an agentic readiness checklist, instruction-file audits, and a workspace context map, and offers a Skill Finder that surfaces repeated prompts so you can turn them into reusable skills.

For teams running the GitHub Copilot app locally, the optional Productivity and Issue credits pages correlate AI sessions with pull requests, merges, and linked GitHub issues.

How engineers can use it

The extension is not on a marketplace, so you build and install the .vsix yourself. The README documents two paths.

Local build with Node.js and npm:

git clone https://github.com/microsoft/ai-engineering-coach.git
cd ai-engineering-coach
npm ci
npm run package

Or open the repo in VS Code, reopen in a Dev Container (Docker or Podman plus the Dev Containers extension), and run npm ci followed by npm run package. Install the resulting file with code --install-extension ai-engineer-coach-*.vsix, or use the Install from VSIX command in the VS Code command palette. After installation, open the palette and run AI Engineer Coach: Open Dashboard, then navigate pages from the sidebar and filter by workspace or harness.

To use it inside the GitHub Copilot app instead, clone the repo, run npm install && npm run build, open the AI Engineer Coach canvas, and reload when prompted.

Documented limitations to note: the token breakdown on the Output page is currently hidden and the Burndown page is disabled; issue credit percentages are explicitly rough relative estimates, not billing figures; and in canvas mode the Skill Finder, Learning Center, Level Up section, and Context Health AI review are hidden because they depend on the local VS Code language model. The README also clarifies this is a community project, not an official Microsoft product.