[Editorial] What the Jev frenzy is really about: agents lack a cheap decision layer

Why it matters

Our take: Jev went viral on X less for inventing classifiers than for productizing the expensive step in agent stacks—routing, scoring, gatekeeping—as a typed decision API. TypeSafe AI (Diogo Almeida, InstructGPT/RLHF lineage) brands this System One: state + questions in, calibrated Choice/Score/Boolean out, no free-form generation. Speed/cost claims and demos (paper tagging, security pipelines, Vercel AI Gateway) sell a stack rewrite; critics say encoders already do this. Both can be true: architectural novelty needs independent checks, but demand for a cheap, calibrated decision layer is already loud. Editorial synthesis, not a lab press release.