Using Jev as a fuzzy linter for AI coding agents: split guidelines into judgeable micro-rules with synthetic evals
mattpocockuk · x · 2026-09-27
Michael Thiessen shares an experiment using Jev as a fuzzy linter that runs after each edit inside an agent harness:
- Converted team coding guidelines into tiny, easily judged rules, keeping only those that don't need extra context/files/reasoning — a single, fast call
- Built a synthetic eval exercising each rule with varied tool calls, plus a held-out set to avoid benchmaxing
- Jev returns confidence scores, enabling downstream handling
Matt Pococku calls it smart: rules like proper useEffect usage or comment hygiene are too complex to lint deterministically yet a waste for full-scale intelligence — a fuzzy linter slots right in between.
More from coding & agent
- Grok Build v1.0.41 ships subagent config inheritance and long-reasoning fixes — mark_k · 2026-09-27
- Harrison Chase on building frontier harnesses: context layers, evals, online learning — VeryWellVersed · 2026-09-27
- Dev shares public Grok Bot 'chief of staff' template running a background specialist agent swarm — RachelVT42 · 2026-09-27
- Meta pitches hand-first dev workflow: one Unity codebase across Quest and VR glasses — Scobleizer · 2026-09-27
- Antigravity 2.0 ships /plan mode as dev mocks it as behind Anthropic's curve — brandon_galang · 2026-09-27
- Codex power users say Astra burns $200 limits in days while Anthropic's cap is hard to hit — brandon_galang · 2026-09-27