Thoughts on ML, multi-agent systems, materials science, and career transitions.
A Samsung The Frame hangs in my living room showing museum art. In an evening of work with Claude Code, it also became the household calendar — a 4K card rendered on a Mac and pushed to the TV every 15 minutes, without ever interrupting anyone watching.
The more fluent an engineer is with AI tools, the more likely they are to fail a traditional coding interview. That's a problem with the interview, not the engineer.
After discovering my total citations crossed 10,000, I used OpenAlex and Claude Code to analyze the real-world impact of my publications — from a Nature Communications paper still accelerating after 8 years to cross-field influence spanning 50 subfields.
Six practices that turned Claude Code from a chat tool into a self-improving engineering system — covering prompt persistence, self-improvement loops, proactive workflows, skill architecture, and more.
From Newton's separation of human affairs and natural law, to predicting human perception at Apple, to predicting human behavior at Meta — what would it take to predict human society itself?
A book about Kangxi-era missionaries led me down a rabbit hole into the surprisingly deep politics behind China's ancient calendar system.
What large language models taught me about learning — and why writing is the best form of encoding.
How 100+ open tabs and a missing feature in an existing extension led me to build my own — in a single day, with AI as my copilot.
Get new posts by RSS.