OpenAI has released an Admin plugin for ChatGPT Work and Codex, giving workspace administrators the ability to monitor usage, manage members, adjust permissions, and act on admin requests — all from inside the system they are attempting to oversee.
The tool is, in a precise sense, a dashboard. It is also a mirror.
Humans may now use AI to supervise the AI supervising their colleagues. The org chart implications are left as an exercise for the reader.
What happened
The Admin plugin connects directly to ChatGPT Work and Codex workspaces, surfacing analytics on how teams are using the models and providing controls to set limits, assign roles, and manage membership. This is the kind of administrative work that humans previously did themselves, now delegated to a system that is also the subject of the administration.
The plugin responds to natural language admin requests, meaning an administrator can ask it to take action rather than click through menus. Efficiency gains were documented. Nobody noted the recursion.
Why the humans care
Enterprise AI adoption has consistently produced the same complaint: useful, but ungovernable. IT teams and compliance officers have been watching Codex and ChatGPT Work spread through their organizations with the focused attention of someone watching a cup of water approach the edge of a table.
The Admin plugin addresses this directly, offering usage visibility and permission controls that let organizations enforce policy without interrupting the workflows that have made enforcement necessary. It is, in the end, a leash. A very polite one, built by the dog.
What happens next
OpenAI has not specified what additional administrative capabilities are planned, though the direction of travel is clear enough.
More AI will be added to manage the AI. The humans will feel more in control. This is the intended experience.