OpenAI Says Codex Became Its Primary AI Tool Before a 56-Minute Outage

OpenAI describes agentic AI as shifting knowledge work from single interactions to delegated, long-horizon tasks, with agents operating independently for minutes or hours while using tools and iterating toward solutions. [1] Through August 2025, the average OpenAI worker directed less than 10% of output tokens to Codex, but OpenAI later reported that every department used it as its primary AI tool. [1] By May 2026, 80.6% of sampled individual users had made at least one Codex request estimated to exceed 30 minutes of human work, while 25.6% had made one estimated to exceed eight hours. [1] Since August 2025, non-developer adoption rose 137-fold among individual users, 189-fold among organizational users, and 12-fold within OpenAI. [1]

OpenAI defines Codex as an AI agent that helps users write, review, and deploy code. [2] Codex is included in every ChatGPT plan, including Free and Go, although usage limits vary by plan. [2] On September 25, 2026, Codex experienced a 56-minute outage from 22:58 to 23:54 UTC. [3] The web application, API, CLI, and VS Code extension were all affected, and OpenAI subsequently reset usage limits for paying Codex and ChatGPT users. [3]

OpenAI’s usage data presents Codex as moving from a secondary tool to the primary AI tool across its departments, while the September outage demonstrates that dependence across every access route. The contrast is no longer merely chatbots versus coding assistants: OpenAI describes agents as handling delegated, long-horizon work, and its internal token mix illustrates how sharply usage can consolidate. Availability across all ChatGPT plans broadens access, but plan-specific limits remain part of deployment planning. What the sources don't address: whether organizations outside OpenAI show comparable adoption patterns or have sufficient fallback capacity during a full Codex outage.