Jev, the decision-only model 20-200x faster than LLMs, spawns 359 projects
Loran750 · x · 2026-09-19
The hottest AI topic this week is #Jev — a non-generative model from TypeSafe AI (Diogo Almeida, co-creator of RLHF) that doesn't produce text. It takes a state (page, ad, ticket, DOM) and returns a structured decision plus a confidence score, running 20-200x faster than a typical LLM at a fraction of the cost, callable directly from code.
Real use cases are piling up: analyzing 724 ads in 40 seconds, driving a browser in real time, classifying posts, blocking ads, routing tickets. The official showcase Jevable lists 359 projects, including:
- TC39 Atlas: scored 324 TC39 proposals across 9 dimensions in 12.7s (25.5 proposals/sec)
- pkg-gate: a 100ms pre-install gate flagging malicious npm lifecycle scripts
- JevForm: forms that dynamically branch on user answers
- traffic-guard: zero-dependency offline anti-scraper gate for reverse proxies
- Jev agents playing Catan — they eventually refuse to negotiate and pass every turn
Skeptics counter that such classification was doable for years with frozen embeddings plus logistic regression; Jev's real edge may be limited to fast-changing categories and calibrated probabilities without a dataset.
Related event: Decision model Jev breaks adoption records amid hype and skepticism(14 posts)→
More from coding & agent
- Devs Lament GitHub's AI Slop: 'A DDoS Coming From Inside the House' — MarkNeumannnn · 2026-09-21
- Dev names understanding the code as the hardest part of AI coding today — BLUECOW009 · 2026-09-21
- GPT cut a process from a full CPU core to 12% in minutes — BLUECOW009 · 2026-09-21
- Jev in production 11 hours after release, saving $127,000 in frontier model tokens — BLUECOW009 · 2026-09-21
- Podcast: AI agent pipelines waste fortunes building a fake human bridge in the middle — thursdai_pod · 2026-09-21
- Indie dev's playbook: go niche, stay disciplined, minimize maintenance debt — menhguin · 2026-09-21