System One and System Two: a metaphor, not model classes

“System One” is useful here as a product-interface metaphor. It is not a verified taxonomy that places every AI model into one of two technical classes. Keep the analogy separate from what the API returns and from what a workload actually achieves.

Jev accuracy in our use is unmeasured because the evaluation was dropped. This conceptual guide supplies no benchmark or finding about model cognition.

Where the contrast comes from

Daniel Kahneman’s Thinking, Fast and Slow contrasts intuitive, rapid thought with more deliberate reasoning. The publisher’s description supports that cognitive contrast. Its account concerns human thought; applying the labels to a software product is an analogy.

TypeSafe’s System One documentation uses the name for a contract in which typed questions over text state produce structured decisions. Inspect that contract to understand Jev’s interface.

What the metaphor can help explain

A reader can distinguish two intended jobs: supplying a proposed answer from an application-defined menu, or producing prose for someone to inspect. That is a useful starting question about the required output.

It is not a finding that one interface is always faster, more accurate or more capable than another. Output form does not establish the serving architecture, model internals or causal explanation for a published price.

Use the primitives guide to name the actual choices: Noul, Choice and Score. Use the published-price guide for charging arithmetic. Neither requires adopting a universal cognitive classification.

Turn the analogy into a concrete question

Instead of asking “Is this a System Two task?”, write what the next step needs. Is it a proposed category, a position on an ordered rubric, an explanation, a draft, or a deterministic comparison?

Then specify the evidence and consequence. A model’s proposed category may still need human review. A prose answer may still be checked by fixed rules. Those are workflow design decisions, not cognitive diagnoses.

For a fictional internal request, a category could help decide which description a reviewer sees first. Whether that helps the reviewer remains a possibility to assess, not an observed Jev result.

Keep product claims and measurements separate

The documented answer shapes establish an API contract. They do not prove general intelligence equivalence, reliable judgment, faster completion or savings for an application.

When discussing this product, use a concrete sentence such as: “TypeSafe documents a bounded answer interface for Jev.” Keep words such as “fast” in their proper context: the book’s metaphor or an explicitly scoped measurement. This element has no such local measurement.

Sources and limits

Sources checked 7 October 2026: Kahneman’s publisher and TypeSafe’s System One contract. The questions and fictional example are original editorial guidance. The metaphor supplies no technical class, architecture, guarantee or implementation result.

For the consumer’s responsibilities, return to bounded decisions and the existing agent-harnesses guide.