Keep exploring · more guides
Jev AI just became the fastest-adopted AI model in the history of Vercel's AI Gateway.
In its first day, close to 13 percent of paying teams were using it, which is double the GPT-5.6 family.
And Jev can't write a single sentence.
It makes the small decisions inside your AI agents in about half a second, and it tells you how sure it is.
I plugged it into my own business this morning and ran over a thousand real decisions for less than one cent.
Stick with me, because I'll show you the three walls I hit setting it up, and the one trick that made the free tier enough.
Vercel runs the gateway, so they can see which models teams switch on. In Jev's first day it reached about 13% of teams — double the GPT-5.6 family and six times Fable 5.1.
They do. They also spend seconds and tokens writing a paragraph every time, when the answer was one word.
Inside one agent run there can be dozens of those small decisions.
What you're watching: my first real call — a lead email goes in, and Jev routes it, scores how urgent it is, and says how likely it is to be a qualified lead.
import { experimental_evaluate as evaluate } from 'ai';
const { answers } = await evaluate({
model: 'typesafe-ai/jev',
state: 'Hi Julian, we run a 12-person dental group in Leeds...',
questions: {
route: { type: 'choice', instructions: 'Which team should handle this?', criteria: {...} },
urgency: { type: 'score', instructions: 'How fast does this need a reply?', criteria: [...] },
qualified: { type: 'boolean', instructions: 'Is this a qualified lead?' },
},
});Inside the AI Profit Boardroom, the agent operating system comes as a zip file. Jev is the decision layer we're adding to it, and the calls are where we set it up with you.
What you're watching: the real setup session, in order — the card error, the blocked model, then Jev answering.
What you're watching: one question against twelve questions about the same email — median of six runs each.
What you're watching: 300 real searches from my Search Console being sorted into how-to, brand, review, what-is and news — a sample of 80 shown.
What you're watching: every page title on agentos.guide being sorted into a topic hub — a sample of 80 shown, 388 in the full run.
AI_InvalidResponseDataError: Question "hub_9" did not select a highest-probability option.
What you're watching: the same pack of ten searches sent to Jev, Claude Haiku 4.5, GPT-5.6 Luna and Claude Sonnet 5 — time first, then cost.
You test it on examples where you already know the answer, then you set your line.
Anything under the line still comes to you.
Members post their wins every day — agency owners, ecom founders, course creators, solo operators across 38 countries.
Read the 158-page wins doc →You saw the three jobs I tested Jev on this morning. Inside the AI Profit Boardroom we set up your version of those, on your leads, your inbox and your content.
