-
Omotenashi, A Week of Noticing in Tokyo
I spent a week in Tokyo for the ClickHouse offsite and I couldn’t stop noticing the small design choices that made li...
-
LLM Benchmarks Are Flatlined. Task Horizons Are Not.
The headline accuracy numbers on standard benchmarks have stagnated. MMLU, TruthfulQA, HellaSwag: the top models have...
-
Ditch grep and Speed up Claude Code with LSPs
Grep is a text search. Code is not text. It’s a graph of symbols, types, and call chains. That gap is where Claude wa...
-
AI B*llsh*tting
Spotting AI Lies: How to Know When Your LLM is BS-ing
-
Edit Survival
Edit Survival - Quality metrics for AI coding agents
-
How To Write A Coding Agent In 169 Lines Of Python
Writing a minimal coding agent from scratch with no hidden magic. Just prompts, tool calls and a loop.