Jev: fast cheap probability models let coding agents self-edit while writing, not after
every · x · 2026-09-26
Jev is a new kind of model: fast, cheap, and accurate enough for judgment calls — feed it text or structured data and it returns a probability (is this email urgent? does this paragraph have a writing tic?).
@hammermt built Jevgram, which scores text against questions like "Does this sound like AI?" His tip: give coding agents specific checks (e.g. spotting "not X but Y" constructions) so they can revise flagged passages as they write rather than after.
The key advantage is cheap enough feedback to run constantly, not just at the end. To start: pick a judgment you'd make in seconds, define what a useful answer looks like, and verify Jev agrees on a handful of examples before running it at scale.
More from coding & agent
- AI generates edge cases, tests and flame graphs: biggest win is team time — remilouf · 2026-09-26
- One prompt with Opus 5.5 + Remotion generates a TanStack AI launch film — HowDevelop · 2026-09-26
- rickasaurus proposes split control-plane and world-input channels for AI agents — rickasaurus · 2026-09-26
- vboykis pushes back on 'languages don't matter': agents need well-designed primitives — vboykis · 2026-09-26
- Dev: AI makes MREs trivial, so he opens PRs with repros instead of issues — remilouf · 2026-09-26
- 60% of DeepSeek Harness users run at least one third-party plugin, team says — vista8 · 2026-09-26