Skip to main content

Meet JEV: A System One AI That Only Decides Without Text

Dragoș Roua introduces JEV, a System One AI designed exclusively for decision-making using structured primitives within an Assess-Decide-Do framework.

AI-written
Inewgen
22 Sep 2026Source: Dev.to3 min read (0 views)
Share
Meet JEV: A System One AI That Only Decides Without Text

Stock photo for illustration only, not from the actual event

Font size
  • JEV is a System One AI model that strictly makes decisions without generating text.
  • It integrates into the Assess – Decide – Do (ADD) framework to build software applications.
  • It answers queries using three pre-defined primitives: noul, choice, and score.
  • Remote API execution took under two minutes, while local implementation took nearly half an hour.

Artificial intelligence development has taken a unique turn with the introduction of a new model named JEV, which breaks away from traditional text-generating systems. Functioning as a System One AI, it focuses solely on making decisions rather than producing output text. Developer Dragoș Roua secured early access and experimented with positioning JEV as the decision-making layer within his Assess – Decide – Do framework, documenting the process in a detailed video demonstration.

JEV cannot output conversational text, but it processes questions using three pre-defined primitives: noul (yes/no), choice (multiple options), and score (probabilities). The creator built a lightweight system capable of generating software based on a single-sentence prompt. JEV forwards this to the Assess stage, where a Large Language Model drafts detailed specifications and tests, following a classic ADD workflow loop.

coding terminal screen computer notebook computer office desk workspace

Stock photo for illustration only, not from the actual event

Once JEV confirms sufficient context, the task moves to the Do phase for code implementation. Upon completion, the process loops back to JEV, which runs the test suites and decides whether the application is finished. The demonstration featured a Python CLI tool designed to count words inside files within a designated folder, highlighted by visual color transitions of red, orange, green, and white flashing circles as tasks migrated between realms.

Never miss the latest news?

Subscribe to get news summaries by email - not often enough to be annoying.

โฆษณา

The application of a System One architectural mindset in AI mirrors rapid, heuristic-based cognitive processing found in human psychology. By restricting JEV to structured outputs instead of natural language generation, the architecture minimizes processing overhead and enhances execution reliability for automated pipeline steps.

The developer conducted two distinct system benchmarks. The first utilized the official remote JEV API alongside a robust OpenRouter LLM, completing the workflow in under two minutes. The second test employed a local JEV implementation paired with a lower-performance local LLM, which required nearly thirty minutes to finish.

2minutes via remote API
30minutes via local setup

This emerging paradigm of structured decision-making models unlocks numerous opportunities for automated software engineering workflows. Developers and enthusiasts are encouraged to share their feedback and potential use cases on the original publication platform.

Source: Dev.to

Comments

Leave a Comment
0/2000

Found something wrong in this article? Report an issue with this article