InstructGPT co-author's Jev harness PDF promises 200x faster, 400x cheaper coding agents
Redo · x · 2026-11-10
- Diogo Almeida, founder of Jev and co-author of InstructGPT, published a PDF on building a Jev harness, claiming up to 200x speed and 400x cost savings.
- The core idea: speed and cost live in the harness around the model — surface hidden decisions inside agents, frame questions a single model answers well, route on calibrated confidence, and hand hard cases to Kimi K3.
- Usage is simple: drop the PDF and article into Claude Code or Codex and let it rebuild its own setup; an evening of work reportedly puts your agent on a harness few teams have.
More from coding & agent
- Medium launches TK, a writing app that plugs your notes into AI agents via MCP — LuoSays · 2026-11-10
- Developer dedicates a Mac mini as a home for his AI agents — ow · 2026-11-10
- Hyo adds Agent Dock to Obsidian: pin and monitor all your AI chats in one sidebar — evielync · 2026-11-10
- GPT 6.1 agent designs a cascadable broadband half-adder in a 23-minute run — CatAstro_Piyush · 2026-11-10
- Claude Code's PM #2 shares how to do devtools PM in the AI era — EricBuess · 2026-11-10
- Dev demos AI that edits documents live while you think out loud — jungofthewon · 2026-11-10