At a glance
WHAT IT’S REALLY ABOUT
Why Jev beats frontier chat models for fast, cheap decisions
- Claire Vo argues Jev is more useful than newer frontier chat models for many workflows because it’s a fast, cheap “decision model” that returns type-safe, predefined values instead of generated prose.
- She explains Jev’s core output primitives—choice, score, and a yes/no-like probability (“null”)—and frames them as the building blocks for routing, triage, and classification at scale.
- She demos high-leverage operational analytics, including clustering thousands of GitHub PRs via pairwise relatedness checks and then labeling clusters with a separate lightweight generative model for initiative reporting.
- She shows personal productivity applications such as analyzing local Claude Code/Codex sessions and Gmail triage, using Jev to filter and bucket items before handing select subsets to more capable models or agents.
- She presents product and app-building patterns, including a multi-model pipeline for a product insights graph and two rapid prototypes: a YouTube comment sentiment/idea dashboard and a voice-to-color emotion mapping app.
IDEAS WORTH REMEMBERING
5 ideasJev is a “decision model”: text in, type-safe values out.
Jev takes unstructured text but returns constrained, predefined outputs (choices, scores, or a yes/no-like probability). That makes it ideal for routing, triage, and clustering where you want dependable structure rather than verbose generation.
Cost and latency unlock large-scale classification workflows.
Because Jev outputs almost nothing, it charges only for input tokens (stated as ~$0.04 per million input tokens) and runs quickly enough for real-time loops. This changes what’s feasible to run at large scale (thousands of items, tens of thousands of comparisons) without major cost.
Pairwise decisions + clustering is a powerful Jev pattern.
She uses Jev for pairwise “are these related?” judgments across PRs, then forms clusters and hands those clusters to a cheap generative model (Gemini Flash-Lite) to label themes. This yields initiative-level reporting (e.g., % effort by area) for pennies.
You can audit your own AI-work patterns using local session logs.
By analyzing locally stored Claude Code/Codex sessions, Jev can categorize what work you’re actually doing over time (engineering vs. agents vs. media/publishing, etc.). The point is meta-insight from data you already have, at negligible cost.
Jev works best as a front-end filter paired with a stronger LLM.
In Gmail triage, Jev quickly scores or tags emails (e.g., safe to delete vs. needs attention), then a second model/agent can do deeper follow-up only on the important buckets. This ‘fast filter + smarter buddy’ setup reduces spend and effort.
WORDS WORTH SAVING
5 quotesNow, what I will say is, more than any other model I've experienced lately, Jev has been the one that has exploded use cases, personal productivity use cases, code use cases, product use cases.
— Claire Vo
Today, I'm gonna give you a very quick whirlwind tour of what Jev is, what it will give you back, and what it won't, and then how I have used it over the last week to do work that I think is worth hundreds of thousands, if not millions of dollars.
— Claire Vo
With Jev, you are getting text in, type-safe values out, and what I mean by type-safe values is these are values that are predefined that then Jev picks from and selects and returns to you.
— Claire Vo
The other thing that you will notice about Jev is it is cheap AF, and it is fast AF.
— Claire Vo
This is one CTOs, VPs of product, V, you know, chief product officers listen up. This is the one that three years ago I would've paid truly $100,000 for.
— Claire Vo
High quality AI-generated summary created from speaker-labeled transcript.
