OpenAI charts Codex taking over its work
OpenAI's new economic paper says Codex is now the primary AI tool in every department, with 80.6% of sampled users running tasks worth over 30 minutes of work.
OpenAI published an economic research paper on June 25 arguing that AI agents, not chatbots, are becoming the main tool for knowledge work, and it used its own staff and Codex users as the evidence.
The headline numbers: by May 2026, 80.6% of sampled individual Codex users had made at least one request the company estimates was worth more than 30 minutes of human work, and 25.6% made one estimated at more than eight hours. Those task-size estimates come from a language model acting as a judge over Codex transcripts, not from timed human work, which is worth keeping in mind. Inside OpenAI, the paper says every department now leans on Codex as its primary AI tool, including non-technical teams like Legal and Recruiting, and non-developers are the fastest-growing group of users.
What it means for you
The direction is more interesting than the exact figures. A coding agent is being pitched as a general work tool that runs long, delegated tasks rather than answering one prompt at a time. Coding agents are already reaching non-developers, and big employers are handing them to whole workforces.
Read the numbers with care. They are self-reported by the vendor and lean on model-estimated task sizes, so treat them as a signal of where work is heading, not settled proof. If you delegate long-horizon tasks to an agent, keep the skills you would lose if the tool went away.