foundations 8 min Updated Sep 19, 2026
System One vs an LLM
Jev selects typed decisions with probabilities. LLMs generate text. Use both — do not force one job onto the other.
Two jobs
| LLM | Jev (System One) | |
|---|---|---|
| Output | Strings / prose / code | Choice, Score, Boolean (+ probabilities) |
| Sampling | Token by token | Parallel answers over declared questions |
| Failure mode | Hallucinated or off-schema text | Cannot leave the schema you defined (product claim) |
| Sweet spot | Writing, planning, tool-argument prose | Route, score, verify, guardrail |
TypeSafe’s launch post frames Jev as a frontier-intelligence function call: unstructured state in, typed probabilistic decisions out.
Composition pattern
- LLM drafts or gathers context.
- Jev decides (queue, score, allow/deny).
- Code branches on confidence.
- LLM writes only when a human-readable artifact is required.