Sep 30, 2026 · 46 min
Listen from 1:16
Listen at 1:16
Jev turns natural language into fast, structured decisions
Jev: 8 real use cases for the fastest, cheapest model I’ve ever used | John Lindquist
The episode shows how constrained classification can make voice interfaces, routing, matching, and agent coordination faster and cheaper than open-ended generation.
- 1Jev maps natural-language requests to functions, routes, scores, and other structured actions instead of generating open-ended text.
- 2Constrained decision spaces make real-time use cases practical, from voice to-do management and deduplication to chess and browser navigation.
- 3Harder tasks still require multi-pass classification, clear categories, strong APIs, or a full generative model for explanation and flexibility.
Don't miss
John Lindquist demonstrates a voice presentation coach that tracks required talking points and converts spoken language into function payloads.
The brief
John Lindquist joins Claire to examine Jev, a fast, inexpensive model that turns natural-language input into structured decisions rather than open-ended prose.
The central idea is to expose the conditional moments inside an application—where code would use if/else or switch statements—and let Jev route language into functions, APIs, and workflows.
Examples span voice to-do management, record deduplication, application-wide command bars, chess, Wikipedia pathfinding, and multi-agent coordination, showing how constrained choices enable rapid exploration.
The standout demonstration is a voice presentation coach that listens for required talking points, checks them off, and turns accumulated speech into function payloads for possible slide control.
Jev is not a universal replacement for generative models: difficult classifications may need multiple passes, while explanation and unconstrained tasks still demand broader models.