I work where technical writing meets AI engineering: making documentation that both people and agents can read, and building the tools that test whether it actually works.
Right now:
- Measuring how documentation format changes the quality of AI-generated code
- Building MCP servers and autonomous agents on top of LLMs
- Scoring API documentation for AI readiness
Selected work:
- Tokens Not Jokin' (Leanpub, 2026), a study of over 21,000 controlled integration tests showing that documentation format explains more variance in AI code generation quality than the choice of model.
- docs-cost-calculator, see what your API docs cost an AI to read. Paste structured documentation and compare token counts across formats.
- analyzeclaudemd, a Flask app that finds common patterns in agent instruction files across GitHub using topic modeling and interactive visualizations.
- pokemon-tcg-mcp, an MCP server that brings Pokémon TCG data to conversational AI, built on the Pokémon TCG API.
- pure-cinema, terminal recording for VS Code. Record, edit, and share sessions locally, fully offline.
- eyeson, a UX analysis tool powered by Gemini that captures multi-viewport screenshots and flags accessibility and AI design anti-patterns.
- Oh My Clawd, a comedy series about a techie who misconfigures things and the eager AI that hallucinates right along with him.
Also building: not-claw, an autonomous agent that uses a Notion MCP server as its backend, and command-line tools that probe and score API docs for AI readiness.
Background: Over 10 years in technical documentation across AWS, Mastercard, and General Dynamics, including API and SDK reference at AWS. I help lead Fort Wayne AI. I like making complex things clear.
📍 Fort Wayne, Indiana | 🔗 grzeti.ch

