Inside OpenAI's Internal Report on 'The Person Who Runs AI 71 Hours a Day': A New Way of Working With Parallel Agents

Drawing on a paper where OpenAI compared its own internal usage data against general and enterprise users, the video walks through a key difference in how AI gets used well. This isn't about rare geniuses or engineers -- it's about how forward-leaning organizations and individuals are actually operating today.

According to the analysis, nearly 30% of people now run five or more AI agents simultaneously instead of one at a time, with the top tier running ten or more at once. The share of individual users delegating a full day's worth of work to AI has grown 10x over the past six months.

The headline figure of '71 hours a day' comes from a simple accounting trick: running time across multiple simultaneous agents gets added together. If one person runs 10 agents at once, that total easily exceeds 24 hours.

The actual paper data shows the top 1% of users went from under 40 hours a day in April to over 70 hours by June. Even the top 5% and 10% are running the equivalent of 10-20 hours simultaneously -- a clear sign that a working style is spreading fast, where humans focus on defining requirements, judgment, oversight, and review while delegating the actual execution to multiple simultaneous agents.

This analysis is based on OpenAI's own internal data, and what stands out is how dramatically usage has shifted from ChatGPT to Codex inside the company -- reportedly at a ratio of roughly 1:99.

This pattern isn't unique to Codex, the video argues -- the same shift is likely to unfold with tools like Claude Code, Cowork, Google's AntiGravity, or Gemini Spark. The shift toward agent-centric work is clearly underway, and it's worth getting comfortable with it early.