greg-isenberg has ryan-vogel (OpenCode founding team) explain Jev, a non-generative decision model created by Diogo Almeida. You pass an input plus an output schema (choices, 0–1 scores), and it returns probabilities in ~200 ms, with no text and no visible reasoning. Demo: 1,700 emails classified by category, priority, spam score, and reply-worthiness for 18 cents (4.2M input tokens). Use cases: lead scoring for a contact form, support triage and routing, instant-quote matching, picking video clips from a transcript, and fast browser control (a flight booked in 7.1 s by Browser Use). It does poorly where real intelligence is needed (a Bitcoin buy/sell signal). Startup lens: find a business with an expensive queue of incoming information and put Jev at the front of it. Access: waitlist, or instantly via the Vercel AI Gateway.
Show: the-startup-ideas-podcast
Concepts: decision-classifiers