For most of the past three years, the race in artificial intelligence has had one direction: make the model talk more, reason longer, and sound more human. So it turned heads this month when a former OpenAI researcher launched an AI model that does the exact opposite. It cannot write a sentence, cannot hold a conversation, and cannot answer an open-ended question. And that, according to its creator, is the whole point.
The model is called JEV, built by a San Francisco startup named TypeSafe AI. The company came out of stealth on September 15 with a $40 million seed round led by DCVC. Its founder, Diogo Almeida, spent years at OpenAI and helped invent reinforcement learning from human feedback, the training method behind ChatGPT. Now he is betting that the next wave of useful AI will not chat at all.
What JEV actually does
Instead of generating paragraphs, JEV returns a decision. You give it a state, which could be a customer message, a document, or even the current position in a game, and you define the kind of answer your software needs. JEV hands back a typed value: a category, a score, or a yes or no, along with the probability that it is right. It does not explain itself in prose. It just makes the call.
TypeSafe describes JEV as a "System One model," a nod to psychologist Daniel Kahneman, who split human thinking into fast, intuitive System 1 judgments and slow, deliberate System 2 reasoning. Large language models like GPT or Claude are System 2 thinkers: powerful, but slow and expensive when you only need a quick answer. JEV is built for the fast lane.
The performance numbers are the reason developers are paying attention. JEV returns answers in 70 to 500 milliseconds and runs, by the company's measure, 40 to 200 times faster than a general chatbot doing the same job. Input costs sit around $0.042 per million tokens, a small fraction of what a frontier model charges. It has already become one of the fastest-adopted launches on the developer platform Vercel, and in a playful demo the model was even wired up to play the classic video game Doom.
Why a decision engine matters for your business
Think about how much of a normal workday is spent on small, repetitive judgments rather than long analysis. A support team decides whether an incoming ticket is urgent. A sales team labels a lead as hot or cold. An operations team approves or rejects a request, or routes each question to the right person. None of these need an essay. They need a fast, cheap, reliable decision, made thousands of times a day.
This is exactly the gap JEV is aiming at. Asking a large chatbot to classify every lead is like hiring a philosophy professor to sort your mail: it works, but it is slow and you are paying for reasoning you never use. A specialized decision model does the same task in milliseconds for a fraction of a cent, which changes the math on automating whole categories of routine work.
The shift toward specialized AI
JEV also points to where the market is heading. The first era of generative AI was about one giant model that could attempt anything. The next era looks more modular, with small, focused models handling specific jobs inside a larger system. Almeida is not alone in this view, and given how many experienced researchers are now leaving the big labs to start companies, JEV is likely the first of many specialized models built for narrow, high-volume tasks.
For teams that want to act on this shift, the practical question is not which single model to adopt, but how to route the right task to the right tool. A decision engine like JEV is powerful only when it is wired into a workflow that feeds it the state, captures its answer, and acts on it automatically. That orchestration layer, connecting fast decision models, larger reasoning models, and your existing software, is where the real productivity gains show up.
For now, JEV is a signal worth watching. It challenges the assumption that better AI always means a smarter conversation. Sometimes the most valuable thing an AI can do is skip the talk and simply decide.
