6 workflows where decision model Jev cuts LLM bills up to 87%
femke_plantinga · x · 2026-10-22
Jev, TypeSafe AI's decision model, replaces the "pick from a list" step in LLM pipelines: instead of generating text, it scores allowed answers and returns a choice, score, or yes/no probability — no output tokens, 70-500ms per decision.
Six workflows: ticket routing (84x cheaper than Opus 5.5 across 77 intents), content moderation (best score of 6 models in one test), AI judges (0.4s per check vs 10s for an LLM judge, per Arize evals), agent safety gates (LangChain's harness already works this way), model routing (only hard requests reach the big model), and docs audit (59 contradictions found in 6 minutes on their help center). Their own runs: cost per run down 87%, twice as fast. Catch: it ranked last on nuanced quality scoring, so benchmark it side by side on a few hundred real cases before switching.
More from coding & agent
- dottxt's ^Ctrl: zero rule violations in 3,508 agent calls vs 41 for the baseline — remilouf · 2026-10-22
- Steve Yegge's 50-agent software factory wants 1:1s and cares about promotions — unsrs · 2026-10-22
- 96.5% deliverability, 42.7% opens: why you can't vibe code a newsletter platform — marclou · 2026-10-22
- New Obsidian plugin turns notes app into an Eagle-style asset manager — jiayuan_jy · 2026-10-22
- One-press Hermes agent: open-source ConcoDispatch turns iPhone Action Button into an AI launcher — Teknium · 2026-10-22
- Scrawl: open-source Mac tool turns screen markup and voice into agent feedback, 680 stars — aronchick · 2026-10-22