When a Good AI Idea Gets Replicated
30 September 2026
A few days ago, I wrote about Jev, a new generation model from TypeSafe AI.
What made Jev interesting was a simple idea:
Instead of asking an LLM to write a long answer and then extracting the decision from it, Jev is built around outputs such as a category, choice or probability.
Almost quietly, OpenAI launched a competing offering at DevDay 2026. It was sandwiched between bigger announcements, so you could almost miss it.
It is called the Decisions API.
The underlying idea is unusually similar: give AI a question with a finite set of possible answers and have it make the decision directly, rather than generate free-form language.
A good idea can create differentiation. But when you sit on top of a powerful foundation model, good ideas can be replicated remarkably quickly. The speed at which the Decisions API appeared is another reminder of just how much leverage foundation-model companies have.
References
