Jev in Practice: Browser Use and Agent Workflows

The author added Jev as a node in an n8n workflow to classify incoming support tickets by solvability, whether it is genuine support, urgency, client and product, at $0.0012 for 35k tokens.

The author says someone rebuilt Claude Code's compaction on Jev and lists five rules before wiring it in: separate deciding from generating, set a confidence threshold, log confidence against real outcomes, and treat self-reported benchmarks as marketing.

The author lists 40 open-source projects that use Jev for structured decisions, including generative UI component selection, a fast Browser Use agent, context compaction, a Monad trading bot, code review pre-filtering, and playing Mario.

The author built an AI scene director with Jev that changes a whole set of Three.js Grassworks scene and grass parameters in real time from a single prompt.

A developer built an amenity icon picker that uses Jev to map the unstructured amenity text landlords write into Phosphor icons for ShaceAi listings.

The DSPy team says Jev popularizing "programming rather than prompting" is good, but hand-writing Jev code is like writing assembly: premature optimization that won't survive the bitter lesson.

Podcast preview: after two years in stealth, Diogo Almeida emerges with $40 million and Jev, discussing how a dedicated decision layer changes agent latency and cost constraints and whether System One splits the market.

The article argues most agents waste tokens on decisions that never needed natural language, and explains how Jev turns routing, scoring and verification into a fast decision layer.

A Replit demo uses Jev to assess space station evacuation risk and pick the drone's next action, with 208 ms average response time and all 12 crew rescued.

A trading bot workflow calls Jev on every new token launch for a GEM or RUG verdict with confidence, runs it in shadow mode to log hit rates before going live, and costs under a cent per scan.

Freek published a post showing how to detect spam and auto-replies using Jev with the Laravel AI SDK, which was called a pretty neat solution.

Asymptote Labs open-sourced Beacon, which records agent activity across 20+ harnesses including Claude Code, Codex and Cursor, and uses Jev to judge which runs are worth learning from and turn them into reusable skills.