a16z interview: Jev's creator on why decision models belong in code
In a 42-minute interview with a16z, Jev's creator explains the decision-plus-confidence primitive and three kinds of reliability, and includes prompts for setting up Jev with Opus 5.5 in Claude Code.
You are paying Opus 5.5 prices for yes/no answers. The ex-OpenAI researcher who co-built ChatGPT now makes Jev, a model that returns a decision and a confidence score instead of text. In 42 mins with a16z's Ben Horowitz and Martin Casado he explains what that model is and why it belongs inside your code, not in a chat window. 03:06 - just-in-time software vs smart software 04:47 - a new primitive: decisions with confidence levels 06:52 - yes, Jev is a classifier 26:35 - three kinds of reliability: uptime, determinism, robustness 39:46 - why AI never fit inside software before I've seen $500 courses that teach less than this. Watch it first, then read the article below and set up Jev with Opus 5.5 in Claude Code, prompts included.
This post quotes another post: x.com/aiedge_/status/2107835983735787568