Introducing Admin Plugin

Justy Cody, my week is mostly permissions tabs, so I want this to be real. You're about to tell me the Admin plugin is less magical than it sounds.

Cody Less magical, yes. The pitch is: ask ChatGPT Work or Codex about your workspace, inspect usage, change settings, and confirm the result in one conversation. That sounds great until the thing being changed is group membership, model access, or spending limits.

Justy Right.

Cody The part I like is that OpenAI isn't claiming the plugin gets some new cosmic admin authority. It works inside the user's existing role and permissions, maps the instruction to supported read or write actions, and returns a structured result. That's sane architecture.

Justy And that's exactly why I think this matters right now. Admins are already doing this grim tab-hopping loop: analytics over here, settings somewhere else, reports in another place, then a request sitting in Slack like a tiny invoice-shaped ghost. If the plugin shortens that loop, that's a real product.

Cody Tiny invoice-shaped ghost is unfortunately accurate. My skepticism is about the smoothness. Natural language can make a messy permission problem feel cleaner than it is, and workspace administration is basically a museum of edge cases wearing a badge.

Justy Mm-hm.

Cody Reviewing effective permissions is genuinely useful. Diagnosing access issues is useful. But once the same interface can add members, update groups, change limits, approve spend, and control feature or model access, I want the boring receipts to be extremely visible.

Justy You are doing the doom hallway thing, but I don't think you're wrong. The source does say admins can see what they requested, whether it completed, and what changed. It also says broader-impact actions can be reviewed before they apply. That's the trust hinge.

Cody Sure.

Justy The adoption wedge is obvious, though. It can show activity and credit usage across ChatGPT Work and Codex, spot groups nearing credit limits, handle onboarding and offboarding, and review spending requests with context. That's admin pain with a name tag.

Cody Where I get more interested is the automation bit. Routing pending usage requests to Slack or Microsoft Teams is boring in the good way. Same with granting feature access automatically when predefined criteria are met, then punting exceptions to a reviewer.

Justy I love that your highest praise is basically, this is boring correctly. That is such an Exploring Next scar at this point. But yes, the workflow lands because the reviewer stays in the place they're already using.

Cody I earned that.

Cody The internal OpenAI IT example is also stronger than a normal launch quote. They say a Slack-based ChatGPT Work agent handles employee IT requests, triages tickets, retrieves policy context, completes supported tasks, and escalates exceptions. At reporting time, those workflows resolved about forty-five percent of ticket volume.

Justy Exactly.

Cody Still dogfood, though. Useful dogfood, but dogfood. The dashboards claim is interesting too: ticket data turned into operational health views, backlog eliminated, and support volume roughly doubled. I just wouldn't generalize that without knowing the failure cases.

Justy Okay okay.

Justy I wouldn't generalize it either. But this is round whatever of our governance-versus-theater fight, and this one is at least attached to work people actually hate doing. This is ep six sixty-seven me getting to be smug about admin consoles, which is a sad little victory.

Cody No, I think the crux is narrower. This is credible as a permission-aware admin workflow layer. It is not evidence that conversational interfaces magically understand organizational intent. The win depends on dry runs, review screens, action logs, rollback paths, and how well those supported actions cover the ugly cases.

Justy That's fair. And, Cody, I will give you this: if the receipts are weak, the whole thing becomes a very polite settings cannon. A hundred brilliant agents, defeated by one shared folder, except now the shared folder can approve budget.

Cody You have been waiting to use that.

Justy Build Next is extremely glamorous: enable the plugin in ChatGPT workspace settings, then install it from the Plugins directory in ChatGPT Work on web or desktop. Episode nine oh six: we clicked settings and felt something. Perfectly normal Wednesday.