OpenAI’s Jev clone could help the frontier lab stop its swarming agents

During OpenAI’s Dev Day event on Tuesday, CEO Sam Altman announced the new “Decisions API,” a tool that provides functionality similar to Jev, a software automation model released by TypeSafe AI earlier this month. Built on a large language model, the Decisions API allows developers to provide a predefined set of options for the company’s Luna model to choose between at high speeds and low costs.

TypeSafe CEO Diogo Almeida reacted on X to the announcement, while noting that OpenAI’s interest could indicate that building in a System One compatible way represents the future. Although OpenAI launched the Decisions API as a limited preview, industry interest is evident in online conversations. Jev-like models and APIs are emerging from various startups, with developers exploring their use in monitoring and securing AI agents cost-effectively.

Cybersecurity professional Shapor Naghibzadeh demonstrated at a recent hackathon that using Jev to verify agentic actions against assigned tasks costs significantly less than utilizing a frontier LLM. Such affordable monitoring could allow these models to run on every agentic action, potentially improving overall agent reliability.