Jev's first two weeks: 29 papers, a wave of clones, and bills dropping from hundreds to a few dollars
The author argues Jev exposed how many click, safety and continue-style decisions in agent loops should never have gone to a chat model, citing 29 arXiv papers and a claim that bills fell from hundreds of dollars to a few.
JEV DIDN'T CREATE A NEW TREND. It exposed a layer everyone was paying a chat model to fake. Two weeks of limited access. 29 papers already on arxiv. A rush of clones. A company brain thread claiming the bill went from hundreds to a few dollars. That's how you know the idea is real. Not because every demo is good. Because people suddenly noticed how much of an agent loop is just: click or don't safe or unsafe this file or that file keep going or stop A chat model can answer those. It just answers them slowly, expensively, and in paragraphs you then have to parse. The interesting part isn't "replace Claude." It's stop waking Claude for work that was never language. Tomorrow the timeline will argue about OpenAI's next agent. That's fine. The quieter question is still the same one Jev forced into the open: how much of your agent is thinking and how much of it is just deciding
This post quotes another post: “Building a company brain with Jev and Opus 5.5: a 10-step architecture”