What is Jev? Inside the new AI model built to make software decisions

Jev is designed to make structured AI decisions for software, focusing on classification, scoring, routing and verification rather than generating conversational responses like conventional LLMs

TypeSafe AI has introduced Jev, its first AI model designed for software decision-making rather than conversation. TypeSafe is an AI company founded by former OpenAI researcher Diogo Almeida along with Erik Gafni and Sasha Sheng. According to TypeSafe, Jev can make quick, structured decisions such as classifying, scoring, routing and verifying information and return the result in a format that software can directly use. The company describes Jev as its first “System One Model”, built for automation rather than conversation. As per the company, Jev is a broader approach to AI systems that can make reliable, machine-readable decisions without first generating a lengthy text response.

 

 

What is Jev and why has TypeSafe built it?

 

According to TypeSafe, Jev was built to help software make quick decisions using AI. The company says current AI models are very good at conversation, but software often needs a simple decision rather than a long answer. Jev is designed to classify information, give a score, choose an option or verify something, and provide the result in a format that software can use directly. 

 

TypeSafe says Jev was inspired by a question its founder, Diogo Almeida, had while working at OpenAI: if AI has become so good at conversation, why is it still difficult to use it for automation inside software? The company named the model after economist William Stanley Jevons and the Jevons paradox, which suggests that making a resource more efficient can sometimes lead to greater use of it.

 

Unlike a regular LLM that typically responds in text, Jev is designed to return structured decisions along with probabilities and confidence levels. According to TypeSafe, this allows software to receive a clear, machine-readable decision that it can act on directly, instead of first interpreting a paragraph of AI-generated text.

 

What does ‘System One Model’ mean?

 

The term comes from psychologist Daniel Kahneman's distinction between System 1 and System 2 thinking. System 1 refers broadly to fast, intuitive judgments, while System 2 is associated with slower, deliberate reasoning.

 

TypeSafe uses the term System One Model for models designed to handle the fast decision-making layer of software. The company says Jev is intended to make many such decisions quickly rather than spend computational resources generating a detailed explanation for every decision.

 

This does not mean Jev is meant to replace the reasoning abilities of larger LLMs. Instead, TypeSafe sees the two approaches as serving different purposes.

 

An LLM could handle a complex task and generate an explanation, while a System One Model could make the smaller decisions needed repeatedly within an automated workflow. 

 

How is Jev different from an LLM such as ChatGPT?

 

The major difference is what the models are designed to produce. A conventional LLM generates text one token at a time. That flexibility allows it to answer a wide range of questions, but TypeSafe says it also creates additional work when the output has to be used by software.

 

Jev does not generate conversational text. Instead, the developer defines the type of decision the model needs to make. TypeSafe's documentation describes three basic decision types: choice, score and yes/no probability. A software application can therefore ask Jev to select an option, assign a score or estimate the probability of a yes/no outcome.

 

For example, an application could provide a customer complaint and ask Jev to determine whether it should be classified as urgent, normal or low priority. Rather than writing an explanation, the model can return the classification and its associated probabilities.

 

TypeSafe says software often needs a decision, not a lengthy AI-generated explanation. Unlike conventional LLMs that return text, Jev produces predefined, structured outputs that can be directly used in software workflows.

 

TypeSafe says this makes the model type-safe, meaning the output follows the structure specified by the software. The company also says Jev cannot produce type errors and does not generate free-form strings. 

 

Where could Jev be used?

 

TypeSafe is targeting software automation where many small decisions have to be made quickly. These could include classification, scoring, routing, screening and AI-agent workflows.

 

For example, an AI system could use Jev to decide whether an incoming request should be escalated, whether content needs additional review, which category a piece of information belongs to, or how likely a particular outcome is.

 

The company says its models are designed to work as an intelligence layer inside software, rather than as an interface that users directly chat with.

 

TypeSafe AI has opened early access for Jev and is bringing developers in from its waitlist. The company is also seeking feedback from developers on where the model works, where it falls short and which software decisions they want to automate.