Jev works best in two steps: build the questions, then run them

ᴅᴀɴɪᴇʟ ᴍɪᴇssʟᴇʀ 🛡️ @DanielMiesslerOriginally posted 2026.10.07

Open original on X

I think most are missing that Jev can/should often be used in a two-step process, not as a single go: 1. Use a super smart model to create a complex and multilayer set of questions using Jev 2. That’s what gets run when “using Jev” In other words you don’t just throw tons of data or hard problems at Jev; what you do is build Jev-based systems that you use for various use cases. Like here’s one for a different way to measure AI coding work: https://workdeeds.ai?v=2

This post quotes another post: “Theo: Jev is a System One model and does not reason”