A few days ago, a brand-new kind of AI quietly shipped. It doesn’t write essays, and it doesn’t write code. I just watched a breakdown from an AI creator who calls it the most exciting AI he’s seen since the early days of ChatGPT, and after seeing the demos, I get it.
The tool is called Jev. The expert explains what it is, what it’s good and bad at, and eight real builds people have already made with it. Here’s my rundown.
What’s new
Jev is basically a very fast decision maker. You give it a question, the rules for “true” and “false,” and some context. It gives you back an answer with a confidence score.
The creator’s favorite example is the classic “Is a hot dog a sandwich?” He sets it up as structured JSON:
- The prompt: “Is a hot dog a sandwich?”
- True criteria: a filling like meat, cheese, vegetables or a spread placed between structural starch
- False criteria: no bread around the filling, only one slice, or a non-bread wrapper like a tortilla, wafer or cookie
- Extra definitions: it even defines what counts as “food”
The result is 73% true. So yes, it’s officially a sandwich.
The pricing is what really got me. Input costs 4.2 cents per million tokens, and output tokens are completely free. You also get $5 in credits to start.
The twist
The magic isn’t how smart each decision is. It’s the speed and the volume. Jev can make thousands, even hundreds of thousands, of choices almost instantly and for almost nothing.
The author uses a great chess analogy. Put Jev against Fable in a normal game of chess and Fable wins. Add a clock, though, and Jev will most likely win, because Fable takes so long to think through each move.
The video is also clear about the limits:
- It’s not a text generator. Don’t ask it how to talk to your boss.
- It’s not good at writing code.
- It works best when you can describe the decision clearly, give solid context and structure the question.
Here’s the clever part. Jev can’t write a web page from scratch. But give it a library of buttons, input fields and fonts, and it’ll put together a whole page in under a second. It picks. It doesn’t create.
The 8 builds that prove it
The original poster pulled most of these from madewithjv.com, a gallery of Jev demos.
- AI slop detector. It scans a site or a piece of writing and flags what looks machine-written. Anthropic’s own site scored 26% slop, and Jev explained exactly which parts set it off.
- Unclutter by Kitze. This one blocks ads and clutter. It strips ads, cookie banners, upsells and annoying pop-ups the moment a page loads. It sends every page element to Jev along with a description of what counts as “spammy,” and Jev decides what goes. You bring your own key, and it costs next to nothing.
- Zapier automations. Jev now works as an action inside Zapier, which sponsored the video. Search for “Jev” and drop it into any workflow across 9,000+ apps. The example in the video accepts or declines calendar invites based on your rules.
- Email priority sorting. Forget newest-first. One build sorted 100 emails by how urgently they need a reply in under half a second.
- Fuzzy find-in-page. A Google product manager built a smarter Ctrl+F. Normal find only matches exact words. With this one, you type roughly what you’re looking for and it highlights the right spot. It’s open source and free.
- Real-time page building. Jev works out what the user needs, like a form or a sign-in box, and builds the page live from ready-made UI parts.
- Clipfast. Type a topic and it cuts a 90+ minute video into matching clips in under 2 seconds. One click and you can download them.
- Color palettes and emojis. Type “80s disco,” “Bowser” or “blue screen of death” and you get a matching color palette. Or type “I need to lose weight” and watch the healthy food emojis float up. It’s just plain fun.
Try it: a mini-workflow
Want to build your first one? Here’s a simple path based on the patterns in the video.
- 🎯 Pick a decision you make over and over. Spam or not. Urgent or not. Accept or decline.
- 📋 Write sharp criteria. This step matters most. Spell out exactly what “true” looks like and what “false” looks like, just like the hot dog example. Define any vague words too.
- ⚡ Feed it volume. Jev pays off at scale. Send it 100 emails, 500 page elements or a full transcript, not just one item.
- 🔁 Act on the answers. Connect the output to an action: hide the element, sort the list, cut the clip. Zapier is the no-code way to do this.
Pro tips
- Give it a menu, not a blank page. If you want Jev to “create” something, give it a library of options to pick from. That’s how the page builder and emoji demos work.
- Use the confidence scores. A 73% “true” isn’t the same as a 99%. Set a cutoff and send the borderline cases to a person or a slower model.
- Pair it with a big model. Let Jev do the fast sorting and filtering, then hand the few items that matter to a model that can reason and write.
- Start with the free credits. At these prices, $5 buys a huge amount of testing, so play with it before you pay anything.
What gets me is how simple each of these builds is. None of them needs a genius model. They need a quick, cheap judge that never gets tired. That opens up a lot of ideas nobody tried before because they were too slow or too expensive.
Watch the full video to see all eight demos running live 🎬