My notes, writeups, and research. I publish when there's something to say, not on a schedule. Each post is its primary location; cross-posts to Habr or elsewhere carry a canonical link back here.
2026
planned Five mis-calibrations engineers make when they pick up LLMs
Teaching four colleagues to use LLMs daily. The over-trust, under-trust, brittle-trust patterns I saw - and what actually built calibration. #llm-adoption#mentoring#calibration
planned A reverse proxy for reading what Claude Code actually sends
What you can learn about system-prompt design from a few thousand intercepted requests. A practical observability layer over an AI coding assistant. #own-router#claude-code#prompt-engineering