Jev Architecture

Aishwarya Srinivasan's 22-minute end-to-end on using System One models inside software, instead of sending every structured check to a text model.

A roughly 10,000-call black-box investigation that reconstructs Jev's likely architecture: a shared state prefix, isolated question branches, and direct probability readouts.

A video walkthrough of the control-system pattern: a request becomes structured state, Jev returns typed decisions, and code keeps the expensive model out of every loop.

A pointer to Diogo Almeida's independent working note on building a Jev harness for coding agents, including why harnesses already make these calls badly.

TypeSafe's case for machine-native composable AI: intelligence as a dependable primitive that software branches on, rather than an assistant that keeps humans in the loop.

Wikipedia's article on Jev: TypeSafe AI, the September 15, 2026 early-access release of jev-1.13.0, the Choice, Score, and Noul primitives, RLCD training, and the unpublished weights and technical paper.

The idea behind Jev: models that return typed decisions and probabilities instead of generated text, named after Kahneman's fast-thinking System 1, with confidence to route around uncertainty.

Why TypeSafe trains decision models with RLCD instead of RLHF: calibrated probabilities where 0.2 outcomes happen about 20% of the time, and the case for machine-to-machine automation.

Instructions, option descriptions, score levels, and noul criteria all accept JSON, so you can label question parts, pass schemas and taxonomies, and keep candidate paths alive.

Break a judgment into atomic Score questions, normalize each by its top level, and combine them with weights you control in code, as in the resume example's four scored dimensions.

An engineering note that places Jev as a typed micro-decision layer inside an agent harness rather than a planner, with preconditions for confidence gates and fan-out and a runtime split into five parts.

A comparative review of Jev and the open-source Laya encoder: interfaces, deployment, disclosed internals, calibration claims, and the limits of agreement-based workflow evals.

Architectural patterns for System One systems: speculative fan-out, confidence-gated routing, composite scoring, and intent routing.

Design AI-powered software by keeping code in control and giving System One narrow, structured decisions to make.