OpenAI is giving workplace administrators a more conversational way to run their AI deployments. Its new Admin plugin for ChatGPT Work and Codex lets authorized IT teams check activity, manage access, adjust limits, and handle routine requests by asking for what they need in a chat instead of jumping between analytics dashboards, settings pages, and reports.

For companies that have moved beyond simply handing employees access to an AI chatbot, that may sound like a small change. In practice, it addresses a very real operational problem: running an AI workspace at scale creates a steady stream of questions about usage, permissions, budgets, onboarding, offboarding, and whether teams are getting value from the tools they have been given.

The Admin plugin is designed to bring those tasks into one place. An administrator might ask which departments are nearing their credit limits, investigate why an employee cannot access a feature, update a group’s permissions, or review a spending request. OpenAI says the plugin can take an admin from the initial question through the supported action and confirmation of the result, all within a single ChatGPT Work or Codex conversation.

That is the bigger idea behind the launch. Rather than positioning AI only as a tool employees use to write, research, code, or analyze data, OpenAI is applying it to the less glamorous work of administering the AI platform itself.

More than a dashboard shortcut

Workplace software administration is usually fragmented by design. Analytics live in one section, identity and member controls live somewhere else, and billing or usage settings may be tucked into another console entirely. That approach offers plenty of controls, but it can make even straightforward requests slower than they need to be.

OpenAI’s pitch is that a conversational interface can act as a front door to those controls. The plugin can help administrators review adoption and credit use across ChatGPT Work and Codex, identify users or groups approaching limits, add or remove members, modify groups, examine effective permissions, and control access to particular features or models based on a person’s role or group. It can also support adjustments to usage limits and help reviewers approve or deny spending requests with the relevant context in view.

The key word, though, is “supported.” This is not meant to be a free-form AI agent with unchecked authority over a company workspace. OpenAI says the plugin operates within the existing role and permission boundaries of the person using it, meaning it does not give an administrator more power than they already have in the underlying system.

That distinction matters. Enterprise IT teams may be open to automating the repetitive parts of management, but they generally do not want an assistant making broad changes based on an ambiguous prompt. OpenAI says the plugin maps a request to a specific, supported read or write action and returns a structured result. For changes with wider impact, admins can review the action before it is applied.

In other words, the company is trying to make administration feel less like navigating a control panel and more like working with an assistant that understands the control panel’s boundaries.

Handling the routine work

The more interesting part of the announcement is the emphasis on recurring workflows. OpenAI says the Admin plugin can automate frequent checks and high-volume requests without requiring companies to build their own internal engineering integrations.

One example is spending requests. Instead of relying on an admin to repeatedly check a queue, the plugin can route pending requests into Slack or Microsoft Teams, where an authorized reviewer can approve or reject them from the collaboration software they already use. OpenAI also says organizations can set criteria for feature-access requests, allowing qualifying requests to be granted automatically while unusual cases are routed to a human for review.

That is a familiar pattern in modern enterprise software: automate the predictable cases, but preserve an escalation path for anything that needs judgment. The value is not necessarily that a language model can decide who deserves access to a particular feature. It is that the model can understand the request, find the relevant policy and workspace context, then move it through an approved process with less manual back-and-forth.

For smaller teams, that could mean fewer administrative distractions for the person who wears the unofficial IT hat. For larger organizations, it could help standardize workflows that otherwise vary depending on which administrator happens to handle a request.

OpenAI frames the plugin as a way to make workspaces more secure, productive, and easier to keep within budget as usage grows. That is a sensible focus, especially as companies experiment with AI access across more departments and face a new version of an old software-management challenge: not just who has a license, but who is using it, how much they are consuming, and whether that usage aligns with company policy.

OpenAI’s own IT use case

OpenAI is also using the release to show how its own IT organization has been applying ChatGPT Work and Codex internally. According to the company, its IT team uses the tools to ask questions about workspace data, generate reports, and move directly from analysis to approved administrative actions.

Kunal Malik, OpenAI’s Head of Global IT, described the value as going beyond faster reporting, arguing that the plugin connects an admin’s question to the next available action, whether that involves checking permissions, changing a group, modifying a usage limit, or reviewing a spending request.

The company also says a ChatGPT Work agent used in Slack now handles parts of employee IT support, including ticket triage, retrieving context, checking approved policies, completing supported tasks, and escalating exceptions. OpenAI says its deployed workflows were resolving about 45 percent of ticket volume at the time it reported the figure.

That number comes from OpenAI’s own account rather than an independent benchmark, so it is best treated as an illustration of the company’s internal deployment rather than a promise of identical results for every organization. Still, it offers a useful view of where this category is headed. AI in the workplace is increasingly being positioned not just as a personal productivity tool, but as an operational layer sitting between employees, systems, policies, and service desks.

OpenAI says its internal IT team has also used ChatGPT Work to turn ticket history and support data into operational dashboards, helping the team eliminate a backlog while support volume roughly doubled. If that workflow translates well outside OpenAI, it could be more significant than the plugin’s ability to change a setting from chat. The real appeal is the possibility of turning messy operational data into a running picture of how an organization is functioning.

A practical step toward agentic administration

There is plenty of hype around “agentic” AI, often involving broad claims that assistants will independently run complex business processes. The Admin plugin is more restrained than that, which may be exactly why it is notable.

It is not asking companies to hand over the keys to an autonomous system. Instead, it wraps existing administrative controls in a conversational layer, keeps permissions in place, and provides confirmation when an action has been completed. For IT departments, that is a much more realistic starting point.

The success of the product will likely depend on how reliably it handles messy real-world requests. Administrators do not always phrase questions neatly, company group structures can be complicated, and access decisions often involve exceptions that are not obvious from a simple policy. But if OpenAI can consistently turn natural-language requests into safe, auditable actions, the plugin could remove a meaningful amount of friction from managing AI workspaces.

The Admin plugin is now available for ChatGPT Work and Codex. To get started, OpenAI says workspace administrators need to enable it in ChatGPT workspace settings and install it through the Plugins directory in the ChatGPT Work web or desktop app.