OpenAI claims 3.1 agent-workdays per human workday, but what does that metric mean?
OliviaYii · reddit · 2026-09-10
OpenAI says its research org now gets 3.1 agent-workdays for every human workday. The poster questions what the metric actually means: does an agent that runs eight hours, needs repeated redirection, and produces half-discarded output still count? They ask multi-agent developers whether runtime matters or only the leftover review-and-fix work.
More from coding & agent
- Full pipeline: niji + GPT Image + local MiniMax H3 + Codex for pixel-perfect animation — Hailuo_AI · 2026-09-10
- Traser: a local deterministic tool to triage agent runs that succeed but return wrong results — Sensitive-Parsnip-12 · 2026-09-10
- Experiment: with a false premise and stale artifacts, only one of three agents found its peers — turtle_bazon · 2026-09-10
- Iron Man fan builds interactive suit teardown site with GPT-6 Astra, Three.js — TheMoonMidas · 2026-09-10
- 8 demos show GPT-6 Astra juggling robots, Portal and 3D on your coffee table — TheMoonMidas · 2026-09-10
- Hermes Agent routing flaw silently switched user to metered OpenRouter, costing $100 — WolframRvnwlf · 2026-09-10