AI Founding Engineer.
Building EdTech products and developer tools.
Building EdTech products at Méthode Aristote to help kids learn.
In my spare time, building open-source tools for developers.
Featured Projects
yt-insights
11Point it at a YouTube channel and get structured insights per video plus an aggregate report. Extracts key takeaways, tools mentioned, and actionable advice. Also scores Short moments with timestamps for clip extraction.
gsc-mcp
9Google Search Console MCP server with 43 tools covering search analytics, SEO intelligence, URL inspection, Indexing API, GA4, CrUX, and schema validation. Includes 9 Claude Code agents.
ctxharness
20AI documentation drift detection. Catches stale facts in CLAUDE.md, AGENTS.md, and .cursorrules before they reach your agents.
StarMapper
82Visualize GitHub stargazers on a world map. No login, just a URL and your GitHub token.
What I Write About
I write long-form articles about making AI-assisted development work in production, not just in demos, using real numbers from real codebases.
New to Claude Code? Start with Where to Start with Claude Code, the onboarding message I have sent on Slack more than 20 times. Then go deeper with Claude Code Under the Hood to understand the 8 tools and the 200K-token context window you are working with.
The Context Engineering series is the backbone of this blog and spans six parts on team systems, portability, and tooling. It opens with The Same Model, Opposite Results (+67% vs -19% on the same stack, with context as the variable).
Production evidence
- AI Velocity Is Bidirectional covers 7 months of data showing debt accumulating as fast as features ship.
- From Afterthought to Infrastructure tracks 506 commits of AI config on a real codebase.
- Non-Technical to Production in 10 Days follows a non-developer modifying 80 files in production with Cursor.
- UVAL is the protocol I built to stop accepting code I don't understand.
Read my career background for the longer story of a VP Engineering building solo with AI agents.
Latest Posts
Your agent config is already 90% portable. The gap that's left fits in one TOML header.
techniqueMoving from Claude Code to Codex to Cursor without rewriting your skills or MCP servers is roughly 90% solved by standards and a bit of generation. The one hard part, converting MCP to Codex TOML with secrets, rulesync does not cover. I read its source to be sure.
Your CLAUDE.md is too long
techniqueYour CLAUDE.md past 200 lines? Two independent Dev With AI talks (Allainmat, Gallet & Dahan) found the same pattern: more rules make Claude less precise, not more.
6/6 · Don't build your moat on one vendor's runtime
techniqueYour AI instruction system is an asset. Locking it to one runtime is a liability. Native primitives, cc-bridge routing, one command to prove portability.