Your AI sent back 137 words. Your code kept one
A category of models that write nothing and return a value in under half a second came into public existence this week. Before you look at it, the test that says how many of your calls never needed a model that writes, and the grid to sort them.
You wired an assistant in to classify your support messages. It works. Except it answers in a paragraph, and someone on your team wrote code to read that paragraph back and pull out the one word you actually wanted. Nobody signed off on that code, it just turned up. On 15 September, TypeSafe shipped Jev, a model that writes nothing and hands back the value directly. Its premise: that paragraph should never have existed. What you walk away with: the exact count of what your product pays for nothing, and the four-question grid to sort your calls. What changes (PM, tech lead, founder): 70 milliseconds against 3 seconds on the same move, and one fragile piece of code gone. Why this week: Jev, the first of these models open to the public, topped Hacker News with 1,920 points. · · · <!--…