Episodebeskrivelse
levelsio said Claude Code edited his whole vlog from one line. so i ran the same line on my own raw footage.
25 minutes of working time, zero fixes from me, and a 10-minute vlog came out. a blind AI judge scored stills from it 6 out of 10. my real edit, with my own agents and about 10 rounds of notes, got 8 out of 10. the gap is direction.
also in this episode: ChatGPT and Codex sharing one limit from December, Firecrawl's headless Clay for agents, your Claude subscription inside OpenCode, agent-desktop, a model editing another model's weights, and the improvements box Patrick Collison put…
Nettsted: share.transistor.fm

