How Agentic AI (Codex) Is Transforming Work at OpenAI

The article exposes the transition from short chatbot interactions to agentic AI that handles long-horizon, independent tasks. At OpenAI, Codex adoption shifted from less than 10% of tokens to over 85% for the average worker, and now accounts for 99.8% of weekly output tokens. This reflects a fundamental change in how knowledge work is done, moving from single interactions to delegated, minutes-long or hours-long tasks.

Concrete internal data shows that by May 2026, 80.6% of individual users made requests estimated to exceed 30 minutes of human work, 70.2% exceeded one hour, and 25.6% exceeded eight hours. Non-developer adoption surged: individual non-developer users rose 137x since August 2025. Departments like Legal, Finance, and Recruiting crossed to Codex as their primary AI tool around April 2026. The heaviest users at OpenAI regularly run over 60 hours of agent work per day across parallel agents.

Agents expand the frontier of what workers can do, enabling non-technical employees to take on coding and technical execution. Over one-fourth of work done by business functions with Codex was engineering or coding. For builders, the key takeaway is that low-friction access to capable agentic tools leads to longer, more complex, cross-functional work, reshaping labor markets and workflow design.