Ten days after Jev launched, the picture of what it is for had already settled: it is a decision engine, and people put it wherever a system used to guess. This page is the map — every use case below links into the corresponding section of the directory, where each entry carries its source and the numbers its author reported.

Triage and routing

The canonical first Jev system: a message arrives, Jev decides where it goes. Support tickets to teams, leads to sales stages, user reports to severity levels. The quickstart example — a ticket routed to billing at 0.84 probability, flagged urgent at 0.999 — is the shape, and builders ship it because it runs inline at 70–500 ms for fractions of a cent. See triage and routing.

Agents and browsers

The largest category by energy. Browser agents use Jev to pick the next action from the page state; coding agents route subtasks; autonomous loops use it as the safety judgment between planning and execution. The pattern name for the split — a language model writes, Jev decides, code acts — is covered in Jev Engineering. See agents and browsers.

Benchmarks, evals, and Jev judging Jev

A surprisingly fast-moving corner: harnesses that score model outputs, compare routing decisions across models, and use Jev as an automated evaluator because judging is exactly what it does. See guides and tutorials for write-ups and the repositories section for the harnesses themselves.

Games and real time

The demo that made the rounds at launch: Jev in a game loop, choosing tactics per tick. Real-time decision loads are the stress test the pricing model invites — many small judgments per second, output free, latency in the hundreds of milliseconds. Builders have it flying drones, playing classic shooters, and simulating transit systems. See games and real time.

Trading and markets

Signal triage, setup classification, news-to-position routing. The interesting constraint here is discipline: Jev proposes the judgment, code enforces the position limits — the layer split doing risk management. See trading and markets.

Content and growth

Scoring, ranking, and gating in content pipelines: which draft matches the brief, which variant ships, which headline fits which audience. The loop is always the same — generate with an LLM, judge with Jev, publish with code. See content and growth.

Research and data

Classification at document scale: sorting thousands of papers or listings by attributes no filter exposes, extraction triage, deduplication judgments. The famous early example — thousands of property listings classified by renovation status and freeway proximity in under 20 seconds for $0.18 — lives in this family. See research and data.

Tools and apps

End-user products where a fast judgment is the feature: classifier playgrounds, routing utilities, command-line tools, browser extensions. See tools and apps.

Robotics and devices

The edge of the map: Jev in the control loop of physical things — drones picking flight decisions, device firmware routing sensor readings, on-device experiments with a remote judge. Latency budgets get tight here, which makes the 70–500 ms window the design constraint builders design around. See robotics and devices.

The pattern behind all of them

Strip the domains away and one architecture remains, which is why this directory sorts by use case instead of by technology: state goes in, typed answers with probabilities come back, code branches on them, and confidence decides what gets a human. If you are deciding where your own idea fits, start with what Jev is, then browse the category closest to your problem and read two or three sources — the receipts are the shortcut to knowing what is real.

Published 25 September 2026. Use-case map reflects the directory’s categories on this date; the listing examples cited are drawn from entries catalogued here with their sources.