{"slug": "what-people-built-with-jev-since-it-launched", "title": "What people built with Jev since it launched", "summary": "TypeSafe AI's System One model Jev has been used by developers to build browser agents, trading bots, and content-analysis pipelines at reported costs as low as $0.0002 per decision and $0.0039 per task. Faadil Shaik reported running 3,282 X posts through Jev for $0.1282 over 8 minutes 34 seconds, while another developer classified 1,018 AI research papers for $0.08 at 256ms median latency per paper. TypeSafe's launch post covers the model, its evals, and its caveats.", "body_md": "Projects, posts and guides about Jev, the System One model from TypeSafe AI. Each entry links to its source and shows the cost and speed its author reported.\n\nBreaking: Browser Use + Jev = Ultrafast ⚡\nFindings flights took 7s and cost only $0.0039 🤯\n> new action space every step\n> DOM state space\n> small LLM fallback to type\n(this video is at 1x speed btw)\nBuilt a tiny open source browser agent. try it below ↓\n\njev is INSANE.\nin 40 seconds it broke down 724 live ads from 37 brands.\nevery hook. every format. offer. cta. awareness stage. landing page mismatch. used 9 cents of tokens.\n(will be avail in @stealads + mcp)\n\nJev + SuperX = virality solved ✅\nEvery post gets 61 questions in ~1s for $0.0004 🤯\n> fitted on 9,481 real posts from 207 creators\n> picks the viral post 2 in 3 times\n> never rewards reply bait\nSo: write, score, rewrite, stop when it peaks.\nFree, no signup. try it below ↓\n\nTypeSafe’s launch post: the model, the evals, and their caveats.\n\nFaadil Shaik\n\n@faadilhshaik\n\ngot @typesafeai’s new model Jev to play Super Mario Bros.\nfast inference + structured outputs makes it surprisingly good for real time use cases.\nI'm excited to see what can be done with these new models!\n\nI gave Jev 3,282 of my X posts across 100M views and asked it to find what actually works for growth.\n4,252,330 tokens $0.1282 for the full 8m 34s run!\nEach post got 8 questions about the topic, hook, tone, whether it teaches something, etc.\nHow-to posts got 150 median likes vs the average median of 44.\nAI and coding was a 1.9x multiplier topic compared and SEO, despite recent posts, was right at base median 1.0x - surprisingly.\nThe recommended topic + angle + voice formula was:\nAI coding + teach something + provocative\n\nI used Jev to classify 1,018 AI research papers.\nThe result: $0.08 total cost and 256ms median end-to-end latency per paper.\nThe pipeline was:\n1. Summarize each paper with DeepSeek V4 Flash\n2. Send the title + summary + 24 possible topics to Jev\n3. Use Jev to classify each paper\n4. Visualize everything on http://1kpapers.com\nThe summaries cost $3.99 on @togethercompute. The classifications cost $0.08 on @typesafeai.\nSo for just over $4 of inference, I ended up with a pretty useful way to explore the top AI research papers from the past year.\nI think this is where things are heading: different models for different parts of the workflow, instead of using one model for everything.\nI’m running evals on the Jev classifications before replacing the current ones, but the site is already live: http://1kpapers.com\n\nAccuracy, cost and time per case for four workflows, by model.\n\nTony Dinh\n\n@tdinh_me\n\nJust trying out Jev, I made a Chrome extension that:\n- Listens to your YouTube audio (optional)\n- Detects if it gets to a sponsor segment\n- Skips it ➡️➡️➡️\n- All in real-time while costing ~$0.005 per video\nPrototype project, BYOK, open-source:\ngithub.com/trungdq88/yout…\n\nJust created this with Jev by @typesafeai. A live viral post analyzer. As soon as you stop typing for .5 seconds it analyzes the viral potential.\nGoing to try and actually make this good, will need to scrape a lot of twitter data...\nNotice how it also categorizes the tweet live... I could have it surface similar tweets on the right side for inspiration... idk just experimenting.\n\nRouting between models, and blocking risky tool calls with a Jev decision.\n\nKyle Jeong\n\n@kylejeong\n\nwe built blazing fast computer/browser use with Jev + @Stagehanddev.\nthis task cost $0.001 and executed at near instant speed (in a remote browser btw)\nthe loop: observe the page, send a11y tree as state + actions as questions, Jev decides the next action, then Stagehand executes it.\n\nComputer use on macOS, with one Jev decision per step.\n\nALAndrew Levin\n\n~$0.0002\n\nJarrod Watts\n\n@jarrodwatts\n\nI built a trading bot with Jev!\nJev decides if it should \"buy\" or \"sell\", given the price feed of an asset pair, and executes real trades.\nIt uses Monad to place the orders on Kuru's on-chain order book in every 300ms block.\nDemo link → jev-trader.vercel.app\n\nwe almost never test new foundation models but we've been testing this for ~a week @every and it's pretty wild.\nthe kind of things that will be obviously indispensible in 6-12 months\nit doesn't produce words as output, it produces probabilities. so it can efficiently act as a judge in cases where you'd need a Fable-level model—but in our testing was 25x faster and 600x lower priced\nexcellent vibe check by @hammer_mt on @every:\nhttps://every.to/also-true-for-humans/mini-vibe-check-typesafe-s-jev-judged-everything-i-ve-written-in-0-7-seconds?utm_cta_source=home_main_a_3\n\nJev picks from the legal moves, compared with reasoning models.\n\nMSMaxim Saplin\n\ntamara\n\n@tamarajtran\n\nfound the perfect use case for @typesafeai Jev:\ninstant compaction\nin 2026, why is compaction still a summarization prompt?\nJev can make it instant by scoring every tool call and dropping what’s irrelevant\n\nJev + Kimi K3 for fraud detection!\nTLDR: Jev classified 100 emails in 1.42 seconds, then I routed the uncertain cases to Kimi K3. The full pipeline got 96/100 correct for only ~$0.07.\nVideo is not sped up, check out the live run!\nHere was my process:\nI gave Jev 100 emails to classify (a mix of 50 legit & 50 fraudelent emails). It classified all of them in 1.42 seconds.\nAn underrated feature about Jev is it will give you the confidence score for a classification, so I routed any prediction under 95% confidence to Kimi K3 to be fully sure.\n31 emails fell below that threshold. After routing those to Kimi K3, the combined pipeline reached 96% accuracy.\nThe full run took 16 seconds & ~$0.07 in inference costs:\n- $0.068 from Kimi K3 on @togethercompute\n- $0.003 (1/3 of a cent) from Jev on @typesafeai.\nI think this is a really interesting pattern: use a fast specialized model like Jev for the narrow task, then route the uncertain cases to a larger LLM.\nI feel like this kind of approach could be a game changer for use cases like fraud or anything realtime. You can use the speed & low cost of Jev while having a larger LLM as a fallback to ensure high accuracy.\n\nThe launch discussion, including the doubts about the benchmarks.\n\nZachi\n\n@iam_zachi\n\nI think I just cooked something 🔥\njev(): a PostgreSQL extension that searches your whole database in natural language. No index, no embeddings, just one function.\nWHERE jev(people, 'could work from home')\nor\nWHERE jev(people, 'name sounds european')\n129 rows judged in ~1s for $0.0009. Second run: 6ms from cache.\n\nOkay so Jev can actually do computer use really well\nWithout any screenshots, or LLMs and no Pixels leave my mac\nI dont even read the Dom elements\nA local CoreML model segments every button and UI element on screen.\nOn-device OCR reads the labels. That text is all Jev gets.\nIt returns a probability across those elements and tells me the best one to click.\nThen it clicks, re-runs detection, and decides again. In a loop until the goal is done.\n~90ms per decision. Faster than any LLM computer use I've tried.\nBlazing fast computer use, without any latency\n@typesafeai is building something really interesting\n\nFull Jev Tutorial\nWhat it is, how you can build with it and what new applications it can unlock\n→ 0:00 Intro\n→ 0:34 Jev explained\n→ 4:06 API setup\n→ 5:59 Demo 1: Voice-controlled browser\n→ 11:33 Demo 2: AI memory\n→ 17:27 Demo 3: YouTube predictor\n\nMoritz Kremb’s video: what Jev is, API setup, and three demos: a voice-controlled browser, AI memory and a YouTube predictor.\n\nIan Nuttall\n\n@iannuttall\n\nCloudflare Workers has Jev now so I'm putting it to the test on keep.md\n- 7x faster search rerank compared to the current hybrid\n- 50x faster tagging of content vs GLM 4.7 Flash with no failures\n\nbuilt `jev-review` @typesafeai\nit's an experimental, local-first MCP plugin that gives coding agents a score quality feedback loop across different metrics.\nagents call jev while they work, get scored, make improvements, and repeat the loop\ntry below 👇\n\ngot @typesafeai's new model Jev as a chief of staff for bots\nJev reads the task, wakes the right teammates off the bench\nand gives each one the right model\nIt is possible on OpenMausBot as it supports all the LLMs from your existing subscriptions\nJev as a decision engine is great\n\nhype-free explanation of jev:\njev does not replace gpt / claude\njev is just a *really* smart switch statement\nlike if 2016 ml classifiers got 2026 levels of intelligence\nit's a new* type of tool that will make a lot of workloads insanely fast, cheap, and accurate\n* = and by new, i mean rebranded\n~~~\nit needs a predefined set of options and it will tell you which one to take\nit cannot:\n- write code\n- generate natural language\n- reason step by step / show its work\n- produce any output you didn't define in advance\n- pick from more than ~255 options in one shot\nbut it can:\n- classify, route, score, rank\n- give confidence\n- pick the right branch, tool, model, or sub-agent\n- judge / verify / guardrail an llm's output\n- label tons and tons of rows\n~~~\ni'd imagine a lot of workflows that look like:\nllm proposes options → jev decides → code executes\nand i see this fitting *really* well with code mode and mcp\n~~~\nimplying this will lead to agi seems incredibly far fetched to me, but i don't want to discount the types of applications that this will make possible\n\nAcabo de terminar la implementación de @typesafeai + Chromium Headless para que mis agentes puedan navegar por internet a una buena velocidad!\nEn este ejemplo le pido al agente que entre a la página del término \"Café\" en Wikipedia y navegue por los hipervínculos hasta terminar en la página \"Inteligencia artificial\"\nLogra completar la prueba en 20 segundos. Creo que pocos humanos podrían hacerlo en ese tiempo.\n\nA browser you control by voice, from Moritz Kremb’s Jev tutorial.\n\nMKMoritz Kremb\n\nPaarangat\n\n@paarangatrai\n\nthis is the easiest way to understand Jev:\nLLMs generate answers.\nJev makes decisions.\nthat sounds like a small difference, but it actually changes the entire use case.\nsay you give a normal LLM this:\n“here’s a user, their account history, payment behavior, support chats, device data, etc.\ntell me if this looks risky.”\nthe LLM might reason through it and return:\n“yes, this looks high risk.”\nmaybe in JSON if you ask nicely.\nwith Jev, you define the possible decisions upfront:\nrisk:\n* low\n* medium\n* high\nmanual review:\n* yes\n* no\nand Jev returns something closer to:\nrisk = high (96%)\nmanual review = yes (91%)\nthat’s basically the product.\nit’s not trying to be another ChatGPT.\nit’s more like an AI-native if statement.\ninstead of:\nif transaction > $10,000:\nreview()\nyou can start thinking more like:\nif “does this behavior look suspicious?” > 95%:\nreview()\nand that opens up a pretty interesting category of software.\na few assumptions I had at first that turned out to be wrong:\n1. “so it’s just a classifier?”\nkind of, but that undersells it.\nthe input can be messy real-world context, and you can ask multiple typed questions about that state at once.\nfraud?\nchurn?\nescalate?\neligible?\npriority?\nall from the same input.\n2. “so it replaces GPT / Claude?”\nnot really.\nI actually think the interesting architecture is:\nJev decides WHAT needs to happen\nClaude / GPT reason or generate WHEN deeper intelligence is needed\nnormal code executes the deterministic stuff.\nJev becomes the routing layer.\n3. “it can’t hallucinate?”\nthis one needs nuance.\nif your allowed answers are:\nLOW\nMEDIUM\nHIGH\nJev won’t suddenly invent:\n“EXTREMELY HIGH 🚨”\nthe output structure is constrained.\nbut it can still be wrong.\nHIGH at 92% can still be the wrong decision.\nso “no hallucinations” doesn’t mean “always correct.”\n4. “why not just force an LLM to return JSON?”\nyou can.\nwe already do this everywhere.\nbut you still deal with generation latency, schema validation, retries, weird outputs, confidence estimation and a lot of glue code.\nJev is designed around the decision itself rather than text generation.\n5. “why should I care?”\nbecause most software is ultimately a giant tree of:\nif this → do that\nif this → route here\nif this → escalate\nif this → reject\nif this → ask a human\nJev is basically asking:\nwhat if those if statements could understand messy human context?\nthat’s a much more interesting framing than “another AI model.”\nI can see this being very useful for:\nfraud / risk\nsupport routing\nmoderation\nPR / QA automation\nlead scoring\ncompliance\nworkflow orchestration\nagent routing\nespecially as the cheap + fast decision layer sitting in front of larger reasoning models.\nearly tech, obviously.\nbut the category itself makes a lot of sense.\n\nPaarangat explains the difference with a user-account example.\n\nnader dabit\n\n@dabit3\n\nAlso have been playing with @typesafeai Jev, insane!\nSo many immediate use cases and new apps are possible. What a time to be a builder!\nSharing some experiments here starting with:\nKeystroke oracle / predictive launcher:\nYour launcher ranks by aliases, fuzzy match, and habit.\nJev reads intent: type \"the pdf I just downloaded\" and the newest PDF is already the top hit with a full confidence on every keystroke, in ~100 ms\n\njev will make agents 10x faster and cheaper, here's how:\n1/ model routing: pick the right model for each task, without training a custom router\nhttps://x.com/mdlahfir/status/2100314182201802811?s=20\n2/ computer use: faster, cheaper and more reliable for action-heavy tasks\nhttps://x.com/gregpr07/status/2100411066966749359\n3/ auto review: ask jev whether an action is safe, instead of using a slow and expensive LLM\nhttps://x.com/fazxes/status/2100300097695232164?s=20\n4/ less obvious: subagent orchestration\nlong-running agents (cursor projects, grokbot, energy) parallelize work with subagents.\nbut every user message, email, or subagent reply can wake the expensive orchestrator.\nexample: it costs $1 to wake up gpt 6 astra w 100k input tokens\njev can decide what each event needs:\n- route directly to a subagent\n- queue for later\n- wake the orchestrator\n\nDavid Fant on model routing, computer use and more, with links to examples.\n\nILIAS ISM\n\n@illyism\n\nNow using @typesafeai Jev in http://aiseotracker.com, http://linkdr.com, http://genppt.com, etc\nAI ends up vibe coding so much AI regex slop if you don't read the code, so I can finally move all this hard-coding to Jev and it's insanely fast!\nAlso for regular LLM calls, it is around 10x faster, 50% cheaper\n\nClaude me integró Jev con Playwright para buscar usados. Lee unos 26 artículos por minuto y decide qué hacer con cada uno en 406 milisegundos\nDescartó los que no encajaban con lo que busco, ofertó por los que sí y hasta les mandó un mensaje a los vendedores cuando faltaba algún dato en la publicación.\nToda la búsqueda salió USD 0,00085. O sea: con un dólar revisás unas 26.000 publicaciones.\nUna IA que por fin puede navegar por internet más rápido que nosotros es un game changer\n\nTL;DR of my new article: WTF is Jev by @typesafeai, and the 9 things people are already building with it. The thesis: 𝗮 𝗰𝗼-𝗰𝗿𝗲𝗮𝘁𝗼𝗿 𝗼𝗳 𝗖𝗵𝗮𝘁𝗚𝗣𝗧 𝘀𝗽𝗲𝗻𝘁 𝘁𝘄𝗼 𝘆𝗲𝗮𝗿𝘀 𝗶𝗻 𝘀𝘁𝗲𝗮𝗹𝘁𝗵 𝗼𝗻 𝗮 𝗺𝗼𝗱𝗲𝗹 𝘁𝗵𝗮𝘁 𝗰𝗮𝗻𝗻𝗼𝘁 𝘄𝗿𝗶𝘁𝗲 𝗮 𝘀𝗲𝗻𝘁𝗲𝗻𝗰𝗲, 𝗮𝗻𝗱 𝗶𝗻𝘀𝗶𝗱𝗲 𝟳𝟮 𝗵𝗼𝘂𝗿𝘀 𝗱𝗲𝘃𝗲𝗹𝗼𝗽𝗲𝗿𝘀 𝘄𝗶𝗿𝗲𝗱 𝗶𝘁 𝗶𝗻𝘁𝗼 𝗲𝘃𝗲𝗿𝘆 𝗰𝗵𝗲𝗮𝗽 𝗷𝘂𝗱𝗴𝗺𝗲𝗻𝘁 𝗰𝗮𝗹𝗹 𝗮𝗻 𝗮𝗴𝗲𝗻𝘁 𝗺𝗮𝗸𝗲𝘀.\nThink AI multiple choice, not AI essay writing. It doesn't chat. You hand it app state plus a typed question, it hands back a decision with a probability attached. 𝟯𝟭.𝟰𝗠 𝘃𝗶𝗲𝘄𝘀 on the launch post in two days (@CompleteSkeptic, who co-invented RLHF). I ran @slashlast30days on it 11 times, then checked every big post by hand.\n🌐 𝗔 𝘁𝗶𝗻𝘆 𝗼𝗽𝗲𝗻 𝘀𝗼𝘂𝗿𝗰𝗲 𝗯𝗿𝗼𝘄𝘀𝗲𝗿 𝗮𝗴𝗲𝗻𝘁 𝗳𝗼𝘂𝗻𝗱 𝗳𝗹𝗶𝗴𝗵𝘁𝘀 𝗶𝗻 𝟳 𝘀𝗲𝗰𝗼𝗻𝗱𝘀 𝗳𝗼𝗿 $𝟬.𝟬𝟬𝟯𝟵. New action space every step, DOM as state, Jev picks the click, a small LLM only wakes up to type. The Browser Use founder built it (@gregpr07, 7.2K likes, 1.8M views) and had to note the video is 1x speed\n🧹 The sleeper: instant compaction. Score every tool call, drop the junk, skip the summarization prompt entirely. \"𝘪𝘯 2026, 𝘸𝘩𝘺 𝘪𝘴 𝘤𝘰𝘮𝘱𝘢𝘤𝘵𝘪𝘰𝘯 𝘴𝘵𝘪𝘭𝘭 𝘢 𝘴𝘶𝘮𝘮𝘢𝘳𝘪𝘻𝘢𝘵𝘪𝘰𝘯 𝘱𝘳𝘰𝘮𝘱𝘵?\" asked @tamarajtran, 5K likes, then shipped the answer that afternoon. Run as a Claude plugin it took a session 𝗳𝗿𝗼𝗺 𝟭𝗠 𝘁𝗼𝗸𝗲𝗻𝘀 𝘁𝗼 𝟴𝟲𝗞 𝗶𝗻 𝗼𝗻𝗲 𝘀𝗲𝗰𝗼𝗻𝗱 (@altryne). Diogo's reply: \"𝘧𝘳𝘦𝘦 𝘤𝘰𝘥𝘪𝘯𝘨 𝘢𝘨𝘦𝘯𝘵𝘴 𝘧𝘳𝘰𝘮 𝘥𝘦𝘴𝘪𝘨𝘯𝘪𝘯𝘨 𝘢𝘳𝘰𝘶𝘯𝘥 𝘵𝘩𝘦 𝘒𝘝 𝘤𝘢𝘤𝘩𝘦\"\n🛡️ Vercel put it in production as the safety reviewer in fx auto mode. 𝗨𝗽 𝘁𝗼 𝟭𝟴𝘅 𝗳𝗮𝘀𝘁𝗲𝗿 𝗮𝘁 𝗽𝟵𝟱 𝗮𝗻𝗱 𝗺𝗼𝗿𝗲 𝗮𝗰𝗰𝘂𝗿𝗮𝘁𝗲 than the model it replaced, per @rauchg, 3.7K likes. LangChain open-sourced the same idea the next day as AutoModeMiddleware. The closed danger classifier inside every coding harness is now a 100ms primitive\n🚦 Model routing as middleware instead of a paragraph in a system prompt. About a dozen lines, probabilities left in agent state so you can audit the choice. The LangChain writeup by @sydneyrunkle is the cleanest how-to-wire-it piece anyone has published\n🔎 RAG precision, solved the dumb way: retrieve as usual, run Jev on every chunk, delete the irrelevant ones. \"𝘢𝘭𝘴𝘰 𝘥𝘪𝘥 𝘢𝘯𝘺𝘰𝘯𝘦 𝘳𝘦𝘢𝘭𝘪𝘻𝘦 𝘫𝘦𝘷 𝘴𝘰𝘭𝘷𝘦𝘥 𝘱𝘳𝘦𝘤𝘪𝘴𝘪𝘰𝘯 𝘪𝘯 𝘙𝘈𝘎?\" (@kushbhuwalka, 416 likes)\n🎮 Minecraft in real time: 𝗝𝗲𝘃 𝗿𝗲𝗮𝗰𝘁𝘀, 𝗚𝗣𝗧-𝟲 𝗔𝘀𝘁𝗿𝗮 𝗽𝗹𝗮𝗻𝘀, and they fight multiple zombies at once (@wuyang_zhou). A launcher that reads intent on every keystroke in about 100ms (@dabit3). TypeSafe's own demo is Doom at 10 decisions a second, roughly $7 an hour\n📬 Email triage at scale: 1,500 emails in batches of 100 with 8 workers, 60,996 views on the demo. \"𝘞𝘦 𝘰𝘯𝘭𝘺 𝘩𝘢𝘷𝘦 𝘢 𝘣𝘢𝘭𝘢𝘯𝘤𝘦 𝘰𝘧 $5 𝘥𝘰𝘸𝘯 𝘩𝘦𝘳𝘦, 𝘸𝘩𝘪𝘤𝘩 𝘫𝘶𝘴𝘵 𝘴𝘩𝘰𝘸𝘴 𝘩𝘰𝘸 𝘤𝘩𝘦𝘢𝘱 𝘵𝘩𝘪𝘴 𝘮𝘰𝘥𝘦𝘭 𝘪𝘴\"\n🗂️ 𝟳𝟳𝟳 𝗷𝘂𝗱𝗴𝗺𝗲𝗻𝘁𝘀 𝗶𝗻 𝘂𝗻𝗱𝗲𝗿 𝟬.𝟳 𝘀𝗲𝗰𝗼𝗻𝗱𝘀 𝗳𝗼𝗿 𝗮 𝗾𝘂𝗮𝗿𝘁𝗲𝗿 𝗼𝗳 𝗮 𝗰𝗲𝗻𝘁. Every's head of evals asked 21 questions of 37 documents in one request, and that is what came back\n🧪 Jev in your browser: Reflex, a Qwen model doing structured decisions on WebGPU, built at Shopify by @kshetrajna and passed around by @tobi. Three independent clones inside 72 hours. 𝗧𝗵𝗲 𝗶𝗻𝘁𝗲𝗿𝗳𝗮𝗰𝗲 𝗶𝘀 𝘁𝗵𝗲 𝗶𝗻𝘃𝗲𝗻𝘁𝗶𝗼𝗻, 𝗻𝗼𝘁 𝘁𝗵𝗲 𝘄𝗲𝗶𝗴𝗵𝘁𝘀\n🔌 Already behind the gateways you use: @vercel AI Gateway inside 48 hours (2,341 likes, the company's second-biggest post), Cloudflare, and @OpenRouter in beta\n💸 𝟱,𝟬𝟬𝟬 𝗿𝗲𝗾𝘂𝗲𝘀𝘁𝘀 𝗳𝗼𝗿 𝗮𝗯𝗼𝘂𝘁 $𝟮. That was one developer counting his bill on day one (@MichaelLee04, 3,060 likes). Input is $0.042 per million tokens. Output is free\n🧨 The honest part: Every's second test came out 𝟮𝟱𝘅 𝗳𝗮𝘀𝘁𝗲𝗿, 𝗻𝗼𝘁 𝟮𝟬𝟬𝘅, and Jev caught 6 of 7 planted defects to Fable 5.1's 7. The HN launch thread (1,863 points) spent most of its length on \"can't hallucinate.\" Top critical comment: \"𝘪𝘵 𝘤𝘢𝘯'𝘵 𝘦𝘮𝘪𝘵 𝘢𝘯 𝘪𝘯𝘷𝘢𝘭𝘪𝘥 𝘵𝘺𝘱𝘦, 𝘣𝘶𝘵 𝘪𝘵 𝘤𝘢𝘯 𝘴𝘵𝘪𝘭𝘭 𝘦𝘮𝘪𝘵 𝘢 𝘤𝘰𝘮𝘱𝘭𝘦𝘵𝘦𝘭𝘺 𝘸𝘳𝘰𝘯𝘨 𝘷𝘢𝘭𝘪𝘥 𝘷𝘢𝘭𝘶𝘦.\" Diogo called the \"it's a zero-shot classifier\" read \"𝘷𝘦𝘳𝘺 𝘢𝘤𝘤𝘶𝘳𝘢𝘵𝘦!\" And the biggest Reddit thread is someone who open-sourced the same architecture a year ago, 1,568 upvotes. Top reply: \"𝘉𝘶𝘵 𝘥𝘪𝘥 𝘺𝘰𝘶 𝘱𝘰𝘴𝘵 𝘪𝘵 𝘴𝘢𝘺𝘪𝘯𝘨 𝘪𝘵'𝘴 𝘵𝘩𝘦 𝘯𝘦𝘹𝘵 𝘣𝘪𝘨 𝘵𝘩𝘪𝘯𝘨? 𝘙𝘰𝘰𝘬𝘪𝘦 𝘮𝘪𝘴𝘵𝘢𝘬𝘦\"\nBonus: the name is not Kahneman. It's William Stanley Jevons, of Jevons paradox. Make a resource cheaper and people consume far more of it. Naming your decision model after that is a thesis statement.\n𝗞𝗲𝗲𝗽 𝘁𝗵𝗲 𝗯𝗶𝗴 𝗺𝗼𝗱𝗲𝗹 𝗳𝗼𝗿 𝘁𝗵𝗲 𝗵𝗮𝗿𝗱 𝘁𝗵𝗶𝗻𝗸𝗶𝗻𝗴 𝗮𝗻𝗱 𝘄𝗿𝗶𝘁𝗶𝗻𝗴. 𝗨𝘀𝗲 𝗝𝗲𝘃 𝗳𝗼𝗿 𝘁𝗵𝗲 𝗿𝗮𝗽𝗶𝗱-𝗳𝗶𝗿𝗲 𝗱𝗲𝗰𝗶𝘀𝗶𝗼𝗻𝘀 𝗶𝗻 𝗯𝗲𝘁𝘄𝗲𝗲𝗻. That's the whole article.\n\nI built a chat bot with jev, no LLM at all! Responses are instant, no hallucinations.\nI hooked it up to web search, wikipedia, weather, todoist and home assistant.\nJev decides what tool to call and what args to use based on the prompt. Instant answers cite sources as well!\n\njev is insane 🤯\nHere is Jev playing subway surfers at super human speed, and also playing 50 games at once.\ncost less than a cent to do this run.\nJev does not replace llms like astra or fable, but opens up an entirely new world of capabilities.\n\njev is insane 🤯\nhere is jev playing smash bros against itself\nhe is controlling all 4 different characters.\nand literally deciding whats the best\nmove to play against itself\nall within a fraction of a second\ni used over 22 million tokens to play this match\nand it only cost me a couple of cents...\njev does not replace gpt6 astra\nbut the possibilities with its instant response time\nare endless\n\nPrediction: millionaires will be made using custom Jev style models (parallel constrained decoding) to make the agent systems companies already run more token efficient.\nLet me explain with a scenario:\nImagine a company already has an agent workflow running where an llm reviews every item before it moves on: a support ticket gets triaged, an invoice gets approved or held, a claim gets flagged.\nEvery one of those goes through a frontier model today, a few seconds and a few cents each, on the way to a decision that in most cases is obvious. Behind that flow sits years of humans (or agents) making the exact same call, with the outcome attached.\nNow imagine you first run each item through a custom PCD or similar model that costs a fraction of the llm and returns a classification of what to do at that step, with a mathematically accurate probability attached.\nWhen it's confident, the item skips the llm entirely.\nWhen it isn't, the llm handles it as normal.\nThe model has seen years of your team making this exact decision, usually a constrained set of decisions, so it should be right most of the time. Say it comes back confident on 6 out of 10 items. That's more than half your llm spend potentially gone from that step, likely with comparable accuracy.\nThis pre processing idea works in a bunch of other use cases too, such as:\n- model/request routing: cheap model, frontier model, or a human\n- picking which skill or subagent to load for a turn instead of stuffing the whole catalog into context\n- reranking retrieved context so only the relevant chunks reach the window\n- guardrails on every agent turn: contradictions, policy issues, prompt injection\n- extracting typed fields from unstructured data emails, PDFs and transcripts before anything expensive touches them\nEvery one of those is a decision an llm makes today, that could potentially be done by another, cheaper model class. Very excited to see Jev/PCD-based pre processing use cases get deployed to agents at scale.\n\nScott Williams on using parallel constrained decoding to make existing agent systems use fewer tokens.\n\nnader dabit\n\n@dabit3\n\nAnother crazy @typesafeai Jev example:\nPredictive spreadsheets\nSpreadsheets recalculate numbers, not meaning. Jev reads intent.\nType \"Urgency\" at the top of a column and, as you type, it figures out you want each row rated from \"no follow-up needed\" to \"urgent\" in ~100 ms.\n\nI build an undetectable realtime adblocker extension with typesafe\nIt checks every dom element and classifies as ad/non-ad and removes it if true\nExtremely fun to work with, expecting an incredible shift in how AI is being used in the future\n\nJev is fun! One-click invoice finder for any website 🧾\n- Automatically finds billing pages using @typesafeai's Jev\n- List/download all invoices with 1 click\n- Works with Stripe billing portals too\n- Remembers where invoices live for next time\nShould I open-source it?\n\nThis is actually insane. This uses @typesafeai Jev model, as a plugin in Claude to review all the un-nesseasary tool calls, and it takes 1s to run!\nLike, literally, 1 second to take my Claude session from nearly 1M to ... 86K tokens! 😮\nAsk your claude to install it and be amazed\nUse this prompt\n```\nInstall, and configure :\nhttps://github.com/tamaratran/fast-jev-compaction\n```\n\nAlex Volkov: a Claude session went from nearly 1M tokens to 86K in about one second.\n\nRaZaan\n\n@razaanstha\n\nI built a Chrome extension for agentic browsing using Jev by @typesafeai, fx.sh including AI Gateway by @vercel.\nNow agents can browse, click, and interact with websites directly in your browser. Cost effective and fassst.\nDecision-making by Jev.\n\nAnnouncing Jev Detector\nThe world's fastest AI slop detector, built on jev from @typesafeai\n~10,000 words scanned for slop in ~2 seconds\nBest part, its free, no sign up required, enjoy!\njevdetector.com\n\nA podcast segment on TypeSafe AI, its $40 million raise, and why the name is a nod to Jevons paradox.\n\nTonino Catapano (tonnoz)\n\n@tonnoz\n\nYou still don't understand the use cases Jev unlocks.\nI've been waiting for something like this since early ChatGPT models.\nprediction: we will see the fastest-growing SaaS by MRR in history within the next month or two", "url": "https://wpnews.pro/news/what-people-built-with-jev-since-it-launched", "canonical_source": "https://madewithjev.com", "published_at": "2026-09-18 18:00:40+00:00", "updated_at": "2026-09-18 18:25:50.057779+00:00", "lang": "en", "topics": ["ai-agents", "large-language-models", "ai-products", "ai-tools", "ai-startups"], "entities": ["TypeSafe AI", "Jev", "Faadil Shaik", "Tony Dinh", "Kyle Jeong", "Stagehand", "Monad", "Kuru"], "alternates": {"html": "https://wpnews.pro/news/what-people-built-with-jev-since-it-launched", "markdown": "https://wpnews.pro/news/what-people-built-with-jev-since-it-launched.md", "text": "https://wpnews.pro/news/what-people-built-with-jev-since-it-launched.txt", "jsonld": "https://wpnews.pro/news/what-people-built-with-jev-since-it-launched.jsonld"}}