Jev in Practice: Browser Use and Agent Workflows

A dev splits 64 papers across 12 Anthropic agents for reading, building and reviewing; the author's demo spider uses Jev for millisecond tool-choice decisions and Opus for the rest.

A course by an Anthropic senior engineer on building self-improving agent teams with loops, graphs and harnesses, plus a full guide to the Opus 5.5 loop and Jev engineering.

Opus 5.5 decomposes tasks into a DAG and signs off, Haiku 5.5 runs subtasks in parallel, and Jev handles micro-branch decisions in milliseconds.

Following Anthropic's advice to call Opus only at key moments, Jev sits one layer down and decides trade or skip and position size for each order.

Repetitive multi-file tasks go to Sonnet 5.5 in parallel with Fable 5.1 cross-checking, while Opus 5.5 handles architecture and reports and Jev absorbs mechanical decisions in 16ms.

A nine-tip manual covers using Jev for routing, complexity scoring, cascade reruns and confidence thresholds so Opus isn't called for every small decision.

The author says Opus 5.5 handles strategy while Jev executes it quickly and cheaply, and calls the combination the most powerful trading bot setup available today.

The post lists four API changes in Opus 5.5 that return 400 errors, such as always-on thinking and removed forced tool use, and says Jev routes effort per turn so the cache stays warm instead of costing 31% more.

Connecting both coding agents to Jev's decision layer routes tasks automatically: in one run it scanned 247 files, sent the security audit to Claude Code and the patch to Codex, passed 18/18 tests and stopped npm publish before execution.

The architecture splits into two layers: Jev handles decisions, priorities and routing, while Opus writes code, researches and calls tools; the author says the bot made $6,000 on its first day and plans to break down the workflow.

A developer wants to use Jev to judge whether Claude will refuse a request and route it to DeepSeek accordingly, but the setup is still just an idea.

The analysis argues that price cuts across Opus 5.5 and GPT-6 tiers will drive AI adoption, while Jev at $0.042 per million input tokens can replace some classification and guardrail calls through structured decisions but does no long-chain reasoning, so its impact on frontier-model demand should be limited.