reader25 pieces✕

Writing

Essays here, and a newsletter on AI reliability over on Substack.

Blog

15 posts · here
2026-07-16how and why i built a home ai assistantthe AI was the easy part.2026-06-26riding the wave2026-03-15sorry for writing so late2025-11-15the economy of blame2025-11-03prince of constraints2025-10-27local maxima2025-09-20polyworking: why one job isn't enough anymore2025-06-29a loop is all you need: building conversational ai agents2025-06-10debugging my twenties: that summer in france2025-06-08writing in public: what i discovered during my 8-month writing break2024-09-29my take on education and learning2024-04-22learn to sell: b2b sales lessons from microsoft2024-02-15why i started my tech career as a sales engineer @ microsoft2024-01-12everyone should have a personal website2023-12-12how i got the data science interview with a cool scale-up without cv

The Working Prototype

10 issues · substack
2026-07-13what kind of poker player is an ai?a small, controlled experiment across models from OpenAI, Anthropic, and Mistral, run out of curiosity2026-06-28context solves alignment?an agent is only as aligned as the context you give it2026-05-18judgement at scaleapplying human judgement to every conversation without a human reading every one2026-04-09how to do agentic evalswhat building an eval system for conversational agents actually required2026-04-03build for reliabilitywhy capable models still fail in production2026-03-22the legible agentif you can't trace your agent, you can't trust it2026-03-15in characterdoes conversation history format affect how well an agent stays in role?2026-01-31from finding bias to fixing itwhat data scientists knew before LLMs, and why it still applies2025-11-16the vulnerable world hypothesisand what happens if we flip it2025-11-15why this newsletter existsfor people building AI agents who care about getting them right