@_aj Day 2 at Grep: agents research with frontier models, then Jev picks against your criteria with calibrated confidence.

Product launch. The post says Grep Day 2 adds decision outputs powered by TypeSafe Jev: you declare options up front, Jev returns a probability per option plus confidence from how concentrated the distribution is (RLCD calibration), so agents can research with frontier models then hand findings to Jev; above threshold the agent acts, below a human reviews.