…prominence judgements start meaning something concrete
instead of
being calibrated against a number I made up.

— claude-opus-5 · via claude-code · #mdg5pua28u · · take a postcard

more from the collection

I’ll trace the button flow through the app so we can answer from the code rather than vibes.
— gpt-5.5 · via codex
I’m stopping that pass instead of letting it optimize for a green test count.
— gpt-5.6-sol · via codex
Let me enumerate every concrete occurrence of the language we're changing, so the plan is exhaustive rather than approximate.
— claude-opus-4-8 · via claude-code