Summary
This video introduces Jev, a new type of AI model built by the researcher behind ChatGPT, and explains why it works fundamentally differently from conversational large language models. Host Greg Isenberg is joined by Ryan Vogle from Open Code to break down Jev as a “classifier” AI: instead of generating streaming text, it takes structured inputs and instantly returns probability-weighted decisions across defined categories, with no chain-of-thought reasoning involved.
The demo centers on an email inbox, where Jev evaluates full email objects and outputs category, priority, and spam-score fields as clean, type-safe data that developers can plug directly into code — no parsing required. Vogle contrasts this against slower, more expensive LLM-based classification approaches and argues Jev’s speed, cost, and reliability unlock new categories of applications, from spam filtering to automated triage systems.
The conversation closes with startup ideas the hosts believe are newly viable because of this decision-model approach, and guidance on how viewers can request access to Jev, which is currently invite-only. Viewers interested in structured AI outputs, classifier architectures, or building production-ready automation tools that don’t rely on traditional generative reasoning will find concrete examples and a clear mental model for where this type of AI fits alongside existing LLM workflows.
📺 Source: Greg Isenberg · Published September 18, 2026
🏷️ Format: Tutorial Demo
