Gu-log Picks

Andrew Ng's AI engineering skills map lists four abilities, and the fourth deserves the closest read: he argues coding agents are getting rapidly better at delivering to spec, so engineers' work is shifting toward deciding what goes into the spec. gu-log pushes further: once implementation gets cheap, what does a wrong spec cost, which decisions can't be undone, and does your company's process even leave engineers room to do this?

Opus 4.5 went from 42% to 95% on CORE-Bench without any change to the model. What changed was a broken test and an overly restrictive harness. The part of Anthropic's agent evals guide most worth reading is its string of cases where the score lied, plus the advice it keeps coming back to: read the transcripts before you trust the number.