Jev in Practice: Browser Use and Agent Workflows

A user who was accepted from the Jev waitlist shortly after applying describes it as making decisions rather than outputting text, claiming it is up to 200x faster and 1000x cheaper and was built by a co-inventor of ChatGPT.

The author predicts companies will use custom Jev-style parallel constrained decoding models to classify tickets, invoices and claims first, skipping the frontier model when confident and cutting LLM costs by around 60%.

A developer released a proof-of-concept MCP for Jev that plugs into existing agent harnesses to verify claims against evidence, screen content before it enters context and rank candidates by meaning.

bitplan gateway lets agents pay per call for inference in Bitcoin SV, including the new Jev model from typesafeai.

Browser Use's Gregor built an open-source browser agent on Jev that found flights in about 7 seconds for roughly $0.0039, using a new action space each step, DOM state, and a small LLM fallback for typing.

A user says using Jev for browser tasks is fun but requires a very different approach than existing models, and shares forked code for deploying your own.

A Stanford CS PhD benchmarks Jev/Typesafe on HiringCafe, his app with 2.5 million monthly active users, scoring resume-to-job-description relevance.

Existing Jev use cases include driving a simulated car, placing trades every 300ms and processing 11,000 legal documents.

The author added Jev support to distilled, using Effect schema to define questions and calling ask directly to get structured data back, calling it badass.

Driveline's R&D team ran basic tests with Jev and found the results promising, but hasn't figured out how to apply it in production, especially around Computer Use.

The author argues that decision models like Jev could have major implications for self-driving, with a video illustrating possible applications.

Jev outputs only Buy/Sell and a confidence score, sending orders straight to Kuru's on-chain order book, with Monad's 300ms blocks driving the trading loop.