🚀 Business & Startups💰 Finance & Crypto
"The new model got worse" is now a monthly ritual. Opus 5 is the latest target - too verbose, over-engineers, treats every bug like a Sev-1.
Half of it is your context. Half is real behavior drift. The problem: your evals can't tell which.
A framework for behavioral evals:...
Aug 19, 2026
