Codex adoption jumps 189x among non-developers as agents take over work

The agent story is becoming measurable in work duration, not just model scores. OpenAI’s new Economic Research paper says Codex is moving from a coding assistant into a broader execution layer for knowledge work, with non-developers adopting it faster than developers in several user groups.

In its June 25, 2026 article, OpenAI reports that by May 2026, 80.6% of sampled individual Codex users had made at least one request estimated to represent more than 30 minutes of human work. The share making a request above one hour reached 70.2%, while 25.6% had assigned at least one task estimated above eight hours. OpenAI notes that these task horizons are estimated by an LLM judge with access to Codex transcripts, so the figures are directional rather than literal time sheets.

The strongest signal is outside engineering. Since August 2025, non-developer Codex users rose 137x among individual users and 189x among organizational users. Inside OpenAI, Legal, Finance, and Recruiting crossed into majority Codex usage around April 2026, after engineering had already moved first. The company says the average OpenAI worker now generates more than 85% of output tokens through Codex, and Codex accounts for 99.8% of weekly output tokens generated inside OpenAI.

The usage pattern also looks less like prompting and more like work orchestration. OpenAI describes agents as systems that can run for minutes or hours, coordinate tool calls, interact with environments, and iterate toward a finished result. Among OpenAI’s daily active internal users, the 99th percentile was generating more than 60 hours of Codex agent turns per day by June 2026, spread across multiple parallel agents.

The caveat is scope: this is OpenAI’s own product telemetry and internal adoption, not a neutral labor-market survey. Still, it gives companies a sharper question to ask. The frontier is no longer only whether AI helps engineers write code faster. It is whether non-specialists can hand off adjacent technical work, data transformation, debugging, and structured analysis that previously required a separate expert queue.