Jev for coding agents — a daily feed

Jevの使い所がだいぶ見えてきた。 失敗パターンも含めて。 「これ意味あるかな？」Claude Codeに入れたJevのプラグインを1週間で外すまで｜shimo4228 https://t.co/zD9bOfyzjz #zenn

@typesafeai very rough and ugly Jev linter https://t.co/9pdx2nWV9f https://t.co/W7vaCoXjHG

テストメーカー（https://t.co/TB57VSv38o）にさっそくJevを導入しました！ AIテスト生成機能でプロンプトを入れている最中に「もっとこれを入れてほしいな」を下に提案してくる機能です。 前々からやりたかったUXだけど低コスト高速のモデルだからこそ実現できた機能。 意外とやってみると考えること多くて難しい💭

Inspired by @brainstormity I build my own little network package checker with Jev @typesafeai ... fun https://t.co/typQhCDD2P

話題のJevをローカルLLMで再現しました。 https://t.co/Cm1wkT9S79 機密情報の見逃し 20件 → 2件 モデルは同じまま。 変えたのは、しきい値の位置だけです。 0.5のままだと、パスワード入りのメモを7割の自信で「問題なし」と通していました、、 https://t.co/3CtiKg6sDZ

itsayush (@itsayush__) Brought Greenwash, a guard that catches coding agents gaming CI, to an IT help desk: KB rules in code, jev for planted instructions and false "done" claims, fail-closed if jev is down, and a live dashboard of every decision. https://t.co/lycBCJtZ2A

and weights aren't the only thing that evolves. /reefine grows the harness itself prompts, skills, tools, rules, config. describe the capability, reef writes it, evaluates it, ships it. the demo adds a /chat mode, a voice notification, and jev as a tool. https://t.co/S00Fh2yLII

@mattpocockuk Hey @mattpocockuk nice use case for Jev! I did a similar thing « Semantic Linting » where you can chose any coding agent to run the lint rules. https://t.co/j5XKKr8RuX Did your AI Hero course btw, was fantastic, made me think about context a lot more!

오늘의 작업 내역. JEV 판독기를 내장시켜서 Opus가 빌드 중에 판정을 확인할 수 있게 패치하였다. JEV의 채점 기준이 될 DB를 만드는게 큰 일이다...어제 스크랩한 소체에 백팩만 따로 붙여보았다. Forge로 넘기기 전 Scrap 수준에서 누리는 막간의 유희. https://t.co/SZBsdi9xXr

6. jevgrep: 905 stars in 2 days. Semantic code search for agent CLIs - ask what code DOES, not what it's named. Jev picks relevance, grep does the rest. Agent tooling keeps eating classic search https://t.co/EuNrnrAzE6

@jevkit/all is at 103 downloads on npm. one install for the whole jevkit ring: record format, linter, drift, benchmark, calibration, vitest. unofficial, not an SDK. npm https://t.co/mMz1gcpgEI built on @typesafeai's jev

@typesafeai Tested on our own large open-sourced repository: https://t.co/kuAF8ovWDF

JEV + HAIKU 5 is outrunning my frontier stack... I took every decision away from the model. It only writes now. Hand it a task → Jev decides what it sees, where it goes, whether it runs → the model types. Full production ship in 3m 40s: → Jev scores the request before anything loads: answerable, in scope, not already done → 31 typed questions run in one call. 0.4s total → It picks the 4 files wort

We solved the “AI writes code” problem. Now we have the “WTF am I supposed to do with this 230-file PR?” problem. Jev Code Reviewer is a new open-source tool that ranks agent changes P0/P1/P2 and shows you what actually deserves attention first. This is the tooling AI coding desperately needs. https://t.co/WsCclVXqSi

I built jev-lint: a fuzzy linter for coding agents. After every edit, a hook asks @typesafeai's Jev whether the new code breaks your team's rules. In 96 Claude Code runs, rule violations per task fell from 2.38 to 1.21. https://t.co/Ti4yRSVykX https://t.co/ozwNp4WENI

Added Jev to Tiny Cyberpunk, now it builds what you say "delete this building", "and this one", "this one will be blue", "arrange those buildings in a shape of letter X", "make all of them green" @typesafeai https://t.co/TA88RqN0cf

jev's repo graph got useful in a hurry 10 more repos built around Jev. zero repeats from the last list commit gates, agent wakeups, test coverage, semantic Postgres, browser cleanup, web search, tax forms, autopilot and Snake jev-commit: checks the staged diff against the commit message, catches debug leftovers and blocks credential-shaped secrets ↳ https://t.co/dAZPj5S8eX SkillRanker: reads a liv

anthropic, openai and everyone else keep dropping better models, nobody's checking how many of those calls even needed to happen went digging inside my own agent loops and found dozens of plain yes or no questions being handed to a model whose whole job is writing paragraphs jev engineering: stop letting a paragraph writer answer yes or no here's the pass that fixed it: ~ scan, surface every hidde

One of the best uses of #Jev so far: ultra fast parallel semantic dimension classification of pull requests https://t.co/SPpCgJfIJB

Jev picked Opus 5.5 at 0.71 confidence to fix 1 Rust bug and the host checked it anyway. The prompt: fix the parser bug in src/parse.rs. keel, the host, builds the menu first. 5 routes. 2 die before anyone chooses. codex · high. Provider not installed. gemini-pro · xhigh. Reasoning level unsupported. Jev gets 3 IDs. No commands. No prose. It scores them: Opus 5.5 medium: 0.71 DeepSeek V4 high: 0.2

opus 5.5 + jev just became a superbrain paste a github repo, a docs site or a product launch page, then press run 34.8 seconds later you know if it's worth building on six channels, one system: 01 input: code, docs, issues, commits, changelog, all at once 02 context: jev maps what's actually there 03 reason: jev fires its typed questions in parallel, maintained or dead, production-ready or demo, r

Why make an AI write an essay when your app just needs a decision? TypeSafe’s JEV makes fast, structured decisions in software. I built JEVRIEL to help AI agents add JEV to new or existing code, then measure the impact. Give your AI JEV wings. https://t.co/zh0kTe2kpt

I've extracted the #PHP driver for interacting with TypeSafe AI #jev from #LLPhant into a stand-alone library, so that it can be used more easily: https://t.co/9rWSoTpNir

I've made a Go pkg for jev, a small, Go client for TypeSafe’s Jev model. The idea: ask AI a question and get a typed result your Go application can use. https://t.co/JC1jObGbkZ

Created a Jev first bug hunter. May be of interest to some. https://t.co/L56hQz6CE5

Built on @typesafeai's Jev (System One). Tracky is open source: https://t.co/P5EerjzBU1

building-with-typesafe-jev: ensina agentes de código a usar decisões tipadas do Jev, com resposta e nível de confiança estruturados Instalação e código disponíveis no repositório https://t.co/2ISnU8McyP

Spent the weekend fine-tuning a DeBERTa v2 model to do Jev-like thing. It's kinda fun and makes me reminisce about my 3rd year in uni doing a Machine Learning unit. https://t.co/wQYh7WgXmR

Introducing jevgrep - a research agent CLI powered by jev from @typesafeai that reduces your coding agent cost by 40% (verified on SWE-bench) Make sure to use the built in skill so your coding agent knows to use jg for context collection https://t.co/BIiicFMlLf

JEV + OPUS 5.5. one prompt, one run, $1.84 → a data science bot. ninety days live: 1,847 questions, $0.22 each, 1,106 answers kept, 276.5 hours returned. then I pointed the same bot at raw event tables. 425 of 1,847. 76% was a number about the warehouse, not the model. one dashboard like this every few days. If this lands, someone on your team already knew it forward it to them. how much of your A

I finally beat Jev with local AI. On a CPU! HP Z2 G1a Ryzen AI + Intern-Decider-4B Took 11 days It’s replacing the Jev API in my dev system AI decisions will be made at home https://t.co/3aP3xurT34

Cloudflare AI(Gemma4+jev)で欲しいアプリ作れたっぽい！ https://t.co/BRTAsWmn06

Instant generative UI with Jev for onboarding 10 different users https://t.co/5hfeGeZN34

Last weekend "I" built a @typesafeai @dotnet SDK and blogged about it 📖 https://t.co/CVDbuZWl6s Today I went further and built a MCP server on top of it 🤩 allowing to ask question like on the screenshot! Blog is coming soon⏳ #dotnet #ai #typesafe https://t.co/cAI31H33Jv

built a coding agent where the model never writes code. Jev only makes decisions from bounded choices. jev-code turns those decisions into ASTs, tools, commands, and actual programs. TypeScript SDK + CLI, fully open source. https://t.co/D3OxFnP4Vs https://t.co/aHdz67UEXd

Last night, I handed over a task to my LangGraph Orchestrated DeepAgent squad to build a simple Next.js crossover of the LangChain and Anthropic websites, featuring a carousel and infinite scroll. [ see the video ] I asked them to create 3 variants and run them unattended. Models used: GPT-6 Luna, JEV, and OpenRouter (free).