Executive Summary
OpenAI has published a research paper highlighting the rapid and widespread adoption of its agentic AI tool, Codex. The analysis shows a fundamental shift in AI usage from short-form chatbot interactions to delegating long-horizon, complex tasks to agents. This trend is prominent both internally at OpenAI, where Codex is now the primary AI tool across all departments, and among external users, with non-developer adoption growing exponentially.
Key Takeaways
* Product Function: Codex is an "agentic AI" that operates independently for minutes or hours to complete complex, long-horizon tasks, a significant evolution from traditional chatbots.
* Widespread Internal Adoption: Codex has become the primary AI tool at OpenAI, accounting for over 85% of output tokens for the average worker and 99.8% of all weekly output tokens generated within the company.
* Usage on Complex Tasks: Users are delegating substantial work to Codex. By May 2026, over 70% of sampled individual users had made a request estimated to exceed one hour of human work, and over 25% made a request exceeding eight hours.
* Rapid Non-Developer Growth: While initially adopted by developers, usage among non-developers has grown much faster. The number of non-developer organizational users has increased 189-fold since August 2025.
* Cross-Functional Capabilities: The tool enables non-technical staff to perform technical tasks. For example, over a quarter of the work done by business function employees at OpenAI using Codex involved coding or engineering.
Strategic Importance
This report signals that agentic AI is moving beyond a niche developer tool to become a core productivity platform for all knowledge workers. By demonstrating massive adoption and value in non-technical roles, OpenAI is positioning its agent technology as the next frontier in enterprise AI, significantly expanding its market beyond simple conversational interfaces.