Jev explained: the 200ms AI classifier that triaged 1,700 emails for $0.18
Jev is a classifier, not a chatbot: feed it an input plus an output schema and get a probability per option in ~200ms. Ryan Vogel demos email triage (1,700 emails for 18 cents), lead scoring, support routing, video clipping and browser control, plus where it fails.
Greg Isenberg sits down with developer Ryan Vogel to explain what Jev, a new kind of AI model, actually is. The one-liner: Jev is a classifier, not a chat model. You define an input (say, an email) and an output schema (is-spam 0–1, category, priority, reply likelihood), and Jev generates no text at all. It returns a probability for each option, like asking what color an iPhone is and getting 80% orange, 10% red, 10% blue. Every query takes about 200 milliseconds regardless of input or output structure.
The headline demo is email triage. Ryan runs Jev over 1,700 of his own emails, scoring each for category, priority, spam score and reply likelihood. The run burns 4.2 million input tokens and 500,000 output tokens and costs 18 cents total. His team stress-tested a $5 intro credit for two days without exhausting it, and he estimates $10 could last around three months. His girlfriend's design agency already uses Jev to score contact-form leads from 0 to 1 so the best leads get a fast human reply.
Greg frames Jev as an AI traffic cop: information comes in, Jev decides what it is, how important it is and what happens next. High-confidence leads go to a human, middling scores go to automation or an LLM, the lowest get ignored. Support routing swaps slow streaming answers for 200ms structured verdicts. The startup angle: find a business with an expensive queue of incoming information and put Jev at the front of it, such as a local-services platform matching "I need my driveway power washed" to the right nearby business and making "instant quote" forms actually instant.
The episode is honest about limits. Asking Jev to buy, hold or sell Bitcoin every minute performs poorly; OpenAI's frontier model does slightly better because it cross-references news. Ryan's advice: use Jev for routing-style decisions and keep it away from your portfolio. Two more demos round it out: a long-video clipper built in about 10 minutes that scores 17 moments in three seconds, and a Browser Use agent that picks a Zurich-to-London flight in 7.1 seconds. To try it today, use the Vercel AI Gateway; direct access is via a waitlist.
FAQ
What is Jev, and how is it different from ChatGPT or Claude?
Jev is a classifier, or decision model, not a chat model. You give it an input and an output schema; it generates no text and returns only a probability for each option (e.g. is-spam 0.9). ChatGPT and Claude produce readable text and reasoning; Jev produces type-safe structured output you can drop straight into code, in about 200 milliseconds per query.
How much does Jev cost? Is it really that cheap?
In the demo, scoring 1,700 emails for category, priority, spam score and reply likelihood used 4.2M input tokens and 500K output tokens and cost 18 cents total. Ryan's team ran a $5 intro credit hard for two days without exhausting it, and he estimates $10 could last about three months.
How do I get access to Jev?
At recording time, direct access to Jev was behind a waitlist. For instant access, call it through the Vercel AI Gateway, where Jev is already available and support has been added to Vercel's AI package. Ryan suggests handing the link to your AI coding agent and letting it wire up the integration.
What is Jev good for, and where does it fall short?
It fits any point where a business makes fast, repeatable decisions on incoming data: email triage, lead scoring 0–1, support ticket routing, auto-picking clips from long videos, choosing options during browser automation. It is not for high-intelligence reasoning: a test asking Jev to buy, hold or sell Bitcoin every minute performed poorly, and Ryan advises keeping it in an advisory, routing role and away from trading.
What startup could you build on Jev?
Greg's framework: find a business with an expensive queue of incoming information and put Jev at the front of it as the sorter. The example in the episode is a local-services platform that matches a request like "I need my driveway power washed" to the best nearby business and turns "instant quote" forms into genuinely instant quotes instead of waiting on a human.