OpenAI Just Published the Number Chip Fabs Stopped Trusting in 1986

OpenAI just said its AI works three point one days… for every ONE day a human researcher works. A chip factory would refuse to put that number on the wall. Because a fab stopped trusting "hours the machine ran" in nineteen eighty-six. On September sixth OpenAI published it: by mid-August, three point one agent-workdays per human workday, and an agent-workday is eight hours of machine runtime. Here's what nobody's telling you. OpenAI said it in the same post: that's a runtime ratio, not a productivity multiplier. In nineteen eighty-six SEMI published E10 and split a tool's day into SIX states, because "running" hid too much. In nineteen ninety-nine, E79 added OEE: availability, times rate, times quality. And the AI world's own data has the missing factors. LinearB measured eight million pull requests this year. AI-written code waits five times longer for a reviewer, and merges thirty-three percent of the time. Human code? Eighty-five. But wait. Agent hours are cheap and elastic. The reviewer is NOT. A fab measures the bottleneck, not the busiest machine. Running is not producing. If you want the number to ask for before your boss shows you the dashboard… follow. Quick tip: for one week, log three numbers per agent run. Minutes it ran, minutes you spent fixing it, shipped or not. Shipped divided by runs is your quality factor. Take that to Monday. Not the hours.