OpenAI published an Economic Research paper in June 2026 documenting how Codex has moved from a developer tool to a primary agentic assistant across the company and beyond. The report measures adoption patterns, task horizons, and token-volume shifts as Codex’s capabilities and product features advanced.
Adoption and usage patterns
According to OpenAI, Codex usage within the company rose sharply over the past year. Through August 2025, less than 10% of tokens used by OpenAI employees were on Codex; by mid-2026 every department, including Legal and Recruiting, used Codex as their primary AI tool. OpenAI reports that Codex accounts for 99.8% of weekly output tokens generated within the company. The firm also cites that 97.9% of 28-day active OpenAI users used Codex, compared with 17.3% for organizations and 0.7% for individuals.
Engineers adopted Codex first; the average engineer shifted the majority of their product usage to Codex by December 2025 and now generates 99% of their output tokens with it. Legal, Finance, and Recruiting crossed into majority Codex usage around April 2026, and the average lawyer or recruiter now generates more than 85% of their output tokens on Codex. OpenAI reports department-level increases in combined output tokens since November 2025: Research median use rose 56x, Customer Support 32x, Engineering 27x, and Legal 13x.
Longer-horizon and non-developer use
OpenAI’s analysis shows a shift from short chatbot interactions to delegated, long-horizon tasks. By May 2026, 80.6% of sampled individual users made at least one Codex request estimated to exceed 30 minutes of human work; 70.2% exceeded one hour, and 25.6% exceeded eight hours. At the 99th percentile of daily active users, Codex agent turns exceeded 60 hours per day by June 2026, distributed across parallel agents, the report states.
Non-developer adoption expanded faster than developer adoption: since August 2025 non-developer individual users rose 137-fold, non-developer organizational users rose 189-fold, and non-developer OpenAI users rose 12-fold, per the report. OpenAI also finds that Codex enables some employees to perform adjacent technical tasks—over one-fourth of Codex work by business-function workers was engineering or coding.
The paper is cited as “The Shift to Agentic AI: Evidence from Codex” (June 2026). OpenAI notes methodology caveats: task horizons were estimated using an LLM-as-judge on a random 0.1% sample of users and the time thresholds are model-estimated and directional rather than exact.
Original source: OpenAI News