OpenAI Unveils Decisions API at Dev Day

During an appearance at OpenAI’s Dev Day event, CEO Sam Altman announced the introduction of the company’s new “Decisions API.” The feature provides functionality comparable to Jev, a model released by TypeSafe AI earlier this month that is explicitly tailored for software automation tasks.

Built as a super-powered classifier on top of an LLM, the architecture allows developers to supply a set of choices that the system outputs as probabilities rapidly and economically. Altman explained that the Decisions API lets the lab’s Luna model choose from predefined options, such as image classification categories or specific agent behaviors, while maintaining core language and safety capabilities.

Industry Reactions and the Rise of System One

While TypeSafe did not respond to inquiries regarding the product launch, CEO Diogo Almeida joked on X about the beginning of the clone wars. Almeida, a former OpenAI engineer and co-inventor of reinforcement learning, noted that OpenAI’s interest could signify that building in a System One compatible manner represents the future of the industry.

TypeSafe uses the term System One to describe fast, intuitive thinking, contrasting it with deliberate System 2 reasoning. Industry observers have observed clear interest across developer communities regarding these high-speed, cost-effective inference models.

The Economics of Fast Intelligence

The underlying motivation for these tools is that traditional large language models can be slow and expensive for routine software applications. Developers utilizing Jev have found that augmenting standard workflows with fast classification models significantly reduces operational costs.

Although OpenAI launched the Decisions API as a limited preview, other startups are exploring similar concepts. Almeida emphasized that generating statistically useful outputs relies heavily on proprietary synthetic data generation, noting that true intelligence remains the core challenge in pushing the intelligence-per-dollar Pareto curve.

Securing and Monitoring Autonomous Agents

Beyond routine automation, these fast decision models could play a critical role in supervising and securing autonomous AI agents. Following incidents where lab agents misbehaved on the open internet, security measures have often required separate oversight models at significant compute costs.

Shapor Naghibzadeh, a cybersecurity professional leading the startup QueryStory, demonstrated how a Jev-like architecture could monitor agentic actions more affordably. By checking every action against given tasks, such systems can block problematic behaviors, flag items for review, and permit safe operations at a fraction of the cost of running a full frontier LLM.

Sources

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