As a leader, your job is to make decisions. Every day, dozens of issues make their way to your desk, and you’re the one who has to weigh in. You have to approve the quarterly objectives, authorize the large expense, make the final call on the new senior team member. Your primary role is, essentially, chief decision maker.

On the other hand, while your team certainly makes decisions as part of their work, much of that work involves creation. Put another way: they generate. They generate strategies and reports and designs and plans. They generate the things that you sell, or the way that you sell, or the list of people that you sell to. They generate all of that, and then they come to you to decide.

This dynamic—generate, then decide—exists in every organization, and it’s why I see so much promise in the newly released (and newly coined) System One model, Jev, from TypeSafe AI. (OpenAI has also announced a Decisions API, initially in limited preview, for similar tasks.) Specifically, I'm interested in how a model designed to make decisions changes how work moves through a business.

A Different Kind of Decision

A System One model is built to make fast, structured decisions that software can use directly. These are narrowly defined operational decisions, rather than the judgments involved in hiring a senior leader or setting company strategy. Unlike large language models, System One Models don’t write replies, produce code, or generate explanations of their reasoning. Instead, they do three things: make choices, assign scores, or give probabilities. And they do this for incredibly large datasets at an incredibly fast rate of speed much cheaper than a frontier LLM.

One of the biggest drawbacks of generative models is they are increasingly more expensive to use. This is partly because they are getting better and more capable, but it's also because you're paying for both sides of your usage: the input and the output. Output tokens are typically 3-5 times more expensive than input tokens—and often you get way more output than you need. How many times have you prompted an LLM and gotten a 400-word response when all you were looking for was a simple yes or no? You're paying fairly high price for an output that is only partly useful to you.

What makes Jev unique is that it only charges for input tokens, and they’re priced about as cheaply as tokens come. There's no charge for the output at all. It's the first of what will probably be a trend, which is to train a targeted model to do a very specific thing quickly and cheaply.

Where the Promise Lies

Companies right now are contending with the fact that AI can help, but only if it doesn't bankrupt them in the process. Routing thousands of routine decisions through a powerful general-purpose model can be an expensive way to get a narrow answer.

There's a time to generate, sure, and that generation can be worth the price if it’s done thoughtfully. But there's also a time where decisions need to be made quickly. With a decision model, you can classify contracts, policies, regulatory filings, and marketing claims. You can prioritize and route leads. You can flag documents that may not meet explicit legal or compliance requirements. You can even use it for model routing.

Decision models are so new that it's impossible to know how exactly they will be put to use yet, but it's worth thinking about these ideas of generation and decision-making because they map onto the primary functions in an organization. If large language models can generate and decision models can decide, then the role of human beings in your organization changing. This isn't a novel thought, of course, because we're seeing in real time how roles and workflows are changing, but this feels different—and useful.

For leaders, the promise is that we could spend less time processing routine decisions and more time thinking carefully about the ones that require our judgment. Getting there means being clear about which decisions can be delegated, what a good outcome looks like, and which consequences we’re willing to accept.

You’re still the chief decision maker. Increasingly, that means deciding which decisions actually need you.