Jev in Practice: Browser Use and Agent Workflows

momito publishes a visual guide to Jev, noting that the guide itself was built with Jev and can be viewed directly in the browser.

A five-step formula: trace the loop, route repeated decisions like tool use, retries, and completion checks through Jev, fall back to the model for uncertain cases, then verify and compare time and cost.

Jev founder and former OpenAI researcher Diogo Almeida, a co-author of ChatGPT and RLHF, argues in a 36-minute video that coding agents like Claude Code are still assistance with a human in the loop, and presents Jev as the next stage: 200x faster, 400x cheaper, zero hallucinations, no human in the loop.

The author says an AI startup has secretly been building a model for controlling programming systems based on his choose-your-adventure calculus formalism, to be presented at Onward! 2026, and links quick experiments using TypeSafe's Jev.

A safety hook uses Jev to judge whether each shell command should be stopped before Claude Code runs it, catching 26 of 27 dangerous commands on a held-out test set with a median decision time of 347ms.

A Chinese guide walks through Jev (TypeSafe System One) from installing the skill and configuring the API key to verifying with curl and sending a real Noul request in Codex, noting it suits classification, filtering and ranking rather than writing code.

The author built an HFT bot on Jev's Choice and Score primitives that filtered 3 real signals out of 1,000 generated in one night, without generating a single word.

Sarah Drasner open-sourced a WebMCP browser extension built on Jev and released a playable shopping cart demo site; the extension is still under review.

The author built a talking board where the server builds a word pack and Jev picks the next word by probability to spell out answers, making constrained choices instead of generating paragraphs.

The author says they helped build the agent battle harness for Los Clankos, driven by both LLMs and TypeSafe's Jev, where agents fight, trade and steal money.

A builder replaced text generation with Jev probability logits so six micro-agents handle invoice disputes, restocking, support and refunds, resolving 1,489 tasks for $14.22 in token cost.

A ranked list of Jev community projects, from MCP integration and context compaction to browser agents and trading bots, ordered from practical infrastructure to experimental.