{
  "id": 9768841,
  "title": "Jev Is Not the Main Story. The Main Story Is How We Spend Intelligence.",
  "url": "https://urgent.news/2026/09/25/jev-is-not-the-main-story-the-main-story-is-how-we-spend-intelligence",
  "topic": "ai",
  "section": "AI",
  "published": "2026-09-25T13:14:03.000Z",
  "source": {
    "name": "Dev.to",
    "slug": "dev-to",
    "url": "https://dev.to/alexitida/jev-is-not-the-main-story-the-main-story-is-how-we-spend-intelligence-2hjd"
  },
  "original_language": "en",
  "account": "Jev may not be the central story, but the way we manage intelligence in AI agents could be far more significant for the future. Traditionally, AI software has been built by sending inputs to large language models for tasks like classification, decision-making, and validation. However, this approach may become problematic as systems expand to include dozens or even hundreds of AI decisions.\n\nTypeSafe AI recently introduced Jev, a model that treats intelligence as a hierarchy rather than a single resource. Jev takes structured questions and state, and returns typed probabilistic decisions that software can consume directly. This approach moves intelligence closer to a probabilistic function call. The architectural shift is substantial, with TypeSafe reporting impressive metrics such as 70-500 milliseconds of latency and $0.042 per million input tokens.\n\nThe author proposes a cognitive governance model consisting of three layers: FAST, REVIEW, and DEEP. FAST layer handles cheap, low-latency judgments, while REVIEW layer deals with uncertain or important decisions requiring more computation. DEEP layer is reserved for expensive reasoning tasks, such as planning and complex diagnosis. This architecture enables the system to allocate intelligence more effectively based on the task's urgency, complexity, and impact.\n\nBy treating intelligence as a hierarchical resource, the system can better manage effort, uncertainty, evidence, and authority. Jev's output is typed, preventing type errors, but it still cannot guarantee 100% accuracy. The distinction between correct and incorrect decisions remains crucial, and the model's probabilistic approach allows software to make better decisions regarding uncertainty.",
  "summary": "TypeSafe's Jev is fascinating. But I think the bigger idea behind it may matter even more for the future of AI agents. For the last few years, the dominant way to build AI software has been surprisingly simple: Send something to a large language model. Need to classify a message? LLM. Need to decide which tool to call? LLM. Need to check whether a result makes sense? LLM. Need to decide whether…",
  "key_points": [
    "Jev model treats intelligence as hierarchical resource instead of single entity",
    "TypeSafe AI reports latency of 70-500 milliseconds and $0.042 per million input tokens",
    "Three-layer cognitive governance model: FAST, REVIEW, and DEEP layers"
  ],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}