The term comes from TypeSafe AI's September 2026 framing for its Jev model, borrowing Daniel Kahneman's System 1 (fast, intuitive judgment) versus System 2 (slow, deliberate reasoning) distinction from behavioral psychology. Because a System One Model's output space is small and enumerated in advance, it can compute a probability over every possible answer at once rather than sampling tokens sequentially, which is the source of both its lower latency and its lack of text-generation ability. It is a complement to, not a replacement for, System 2-style reasoning LLMs — the pattern is using an LLM to define a decision, then routing the repeated structured part of it through a System One Model.