OpenAI says Codex agents now handle longer cross-functional work internally

Codex as an internal work system

OpenAI’s post is less about a new model name and more about how agentic tools are being used inside a frontier AI lab. In a June 25, 2026 tweet at 17:23:08 UTC, OpenAI said agents are transforming work in every department and that people across the company use Codex for more complex, longer-running, and increasingly cross-functional tasks. FxTwitter showed about 1.12 million views, more than 5,200 likes, and over 1,500 bookmarks during collection.

“more complex, longer-running, and increasingly cross-functional.”

OpenAI’s official account typically publishes product releases, research updates, safety notes, and selected internal-use stories. This one is material because it frames Codex as part of company-wide workflow rather than only a developer assistant. Longer-running agent work requires more than code completion. It needs task decomposition, file and document access, tests, review checkpoints, and enough context to move between engineering, operations, support, policy, and product concerns.

The post does not provide a productivity benchmark, so the claim should be read as evidence of adoption rather than proof of measured output gains. The concrete numbers are engagement and scope: more than one million views for a post about internal agent use, and language that points to every department rather than a single engineering team. That distinction matters for enterprise buyers. If the same patterns become productized, the competitive question shifts from which model writes the best function to which platform can safely run work across many internal systems.

What to watch next is whether OpenAI exposes the internal patterns as Codex features. Long-running work needs persistent execution, permission boundaries, logs, human approval, and failure recovery. If those controls reach ChatGPT, the API, or enterprise Codex deployments, agent tools will look less like chat assistants and more like managed work queues for software and business operations. Source: OpenAI source tweet