Jev's five-step decision loop: LLM plans, Jev picks the route
TypeSafe co-founder Diogo Almeida explains that Jev does not do conversational intelligence but outputs deterministic choices at branch points, paired with LLM analysis, tool execution and state updates, making it many times faster and hundreds of times cheaper than classic LLM queries.
I am genuinely surprised why everyone isn't using this yet Diogo Almeida, co-founder of TypeSafe AI, which developed the Jev system, and previously at OpenAI and Google Brain, stated a simple idea: "The point of Jev is not to make the model smarter in a conversational sense, but to make it completely deterministic at decision points. If you have a choice of N actions, the system should produce a precise choice without hallucinations or extra tokens" 1. Analysis and reasoning: A large language model (LLM) performs the initial deep analysis of the task and creates an overall plan of action 2. Decision filtering: At a branching point, a specialized layer (Jev) instantly selects the optimal route from the available options without calling an expensive LLM 3. Execution: External tools or sub-agents perform the selected action in the environment 4. State update: The system captures new data and updates the current context 5. Re-selection: Jev immediately makes the next local decision based on the new state, closing the loop As a result, you get a ready-made, highly accurate decision-making solution for agents that works many times faster and costs hundreds of times cheaper than classic LLM queries More useful projects in the article below
This post quotes another post: x.com/bober_smart/status/2103462864870686848