Jev public demos collected: track cost per completed task
A roundup of public Jev demos — Browser Use flight search, classifying 1,018 papers, sorting 500 emails, cutting a Claude session to 86K tokens — arguing the metric to track is cost per completed task, not per decision.
THIS IS PURE F*CKING GOLD: every public Jev demo so far, as one receipt. the fine print matters more than the price > Browser Use found flights in 7 seconds for $0.0039 > nutlope classified 1,018 research papers for $0.08 total, 256ms per paper > rileybrown sorted 500 emails in seconds for 3.5 cents > a Jev plugin cut a Claude session from ~1M to 86K tokens in about a second the fine print the article actually spells out: > the flight agent finds results. it doesn't book tickets > the 7 seconds start after the first page read and skip the final check > $0.042 per million input tokens is the price of the decision, not the whole run > the Browser Use speedup came from cutting browser calls from 1,092 to 101, same models so the number to track isn't price per call. it's cost per completed task. a cheap decision that sends a worker down the wrong branch costs more than the decision the video is that receipt paste this into Claude Code ↓ "Log every agent run as one row: task, model calls, decision calls, tool calls, total cost, finished yes/no. Flag runs that cost more than [X] or didn't finish. For each flagged run, show the first wrong branch." the full 10-step roadmap is in the article below what's the cheapest wrong branch that cost you the most?
This post quotes another post: “Long read: how to build a Jev-powered agent stack”