In our experience, AI rollouts often start the same way. Someone picks a plan, buys the seats, and sends everyone a login. On the project plan, that looks like the work. A few months later the questions start. Why can some people use a feature and others can't? Who turned that connector on? What did we agree to share with this vendor?
The purchase is the smaller part of the project. The larger part is configuration: the settings that decide who can use what, what the AI can see, and what it is allowed to do. The vendors' own documentation shows how much there is to configure. And the settings keep moving.
The setup is bigger than the purchase.
Start with the decisions that come before anyone logs in. A Claude Team plan has two seat types with different usage allowances, and the limits apply to each person rather than to the team. Anthropic's help center says an organization can turn on usage credits so people can keep working after they hit their limit. Someone has to decide who gets which seat, and whether to turn those credits on.
Microsoft splits the work across more than one place. Copilot licenses and agent availability are managed in the Microsoft 365 admin center. Agents built in Copilot Studio are governed in the Power Platform admin center, and building them takes a license for the organization plus a license for each person who builds. People who only use a published agent need no special license. Microsoft's own deployment playbook recommends starting with a small pilot group and assigning licenses through groups in Entra ID, so that adding and removing people doesn't become a manual job.
OpenAI's workspaces differ by plan. In ChatGPT Business, apps are on by default, and an admin can change availability for the whole workspace. In Enterprise and Edu, new plugins and apps are generally off by default, and admins can give access to specific roles. Role-specific app access applies to Enterprise and Edu. On Business, admins decide whether an app is on for the whole workspace. And installing a plugin doesn't give anyone access to the apps inside it. Those have to be allowed separately.
None of these decisions is hard on its own. There are just a lot of them, they live in different admin screens, and the right answer depends on how your teams work, not on what the pricing page suggests.
The settings keep changing.
Read the vendors' announcements for a month and the pattern is clear. On October 8, Anthropic took Docs, Slides, and Design out of beta on every Claude plan. For Enterprise organizations they turned on by default on October 15, though admins could turn them on earlier. Dashboards and Motion launched in beta the same day, and for Enterprise they are off by default until an admin turns them on. Sharing slides and dashboards with people outside the company works only if the admin allows it.
On November 10, Anthropic's interactive connectors began opening as full apps beside the chat, with Claude able to edit the open page, row, or file. The announcement tells admins that if their organization already allows all of a connector's tools, the new features appear as they roll out. A choice someone made months ago about a Notion or Confluence connector can now decide whether Claude can change things there.
Microsoft changed how Copilot is sold. Since October 16, the existing Copilot offers that require a Microsoft 365 base subscription carry "(add-on)" in their names. Starting December 1, new Microsoft 365 Copilot purchases move to offers that don't require a base subscription, and the add-on offers reach end of sale for partners, though existing customers can keep renewing. Microsoft says the capabilities aren't changing, only how customers buy them. Anyone with a purchasing process still has to keep up with the names.
OpenAI builds this into its own advice. Its Enterprise admin quickstart says that "ChatGPT regularly adds new features, capabilities, and ways to connect with external tools," and that some of them "may require admin decisions about data access, sharing, external services, or write actions." It tells admins to revisit their settings regularly as features are added.
That is two vendors changing what an administrator has to decide in a single season, and a third telling administrators to expect more of the same.
Large and small companies feel it differently.
A large company feels the configuration work as a stack of reviews: vendor conversations, contracts, data handling, identity and access. A smaller company feels it differently. The same AI products and the same settings come with fewer people to work through them. McKinsey's 2025 survey found nearly half of respondents at the largest companies had reached the scaling phase with AI, compared with 29 percent of respondents at companies in the lowest revenue band. McKinsey doesn't say why, and configuration is only one possible reason. A company without a dedicated IT team still has the same setup to do with fewer hands.
What to do before anyone signs in.
The same few moves help at any size. Name one owner for the configuration, someone with the authority to make the calls rather than someone squeezed in between other work. Decide the questions that matter before launch: which AI products and plans, who gets a seat, which data and connectors are in bounds, and what the AI can do without asking. Start with a pilot group so you can see how the settings behave with real people. Then put a recurring review on the calendar, because the vendors will keep changing the defaults and each change is a decision.
Most of this is ordinary administration. It just has to be done by someone who knows what the settings mean.
Where we come in.
We now offer this as a service called Outsourced Expertise. For a couple of months, we put someone who has done these rollouts before into your building. They help choose the plans, work through procurement, and set up the workspaces, connectors, permissions, and admin controls before your people sign in, for Claude, Cowork, ChatGPT, Microsoft Copilot, and Copilot Studio. The details are on our services page.
Sources
- Anthropic, What is the Team plan? (Claude Help Center: seat types, per-member weekly limits, usage credits)
- Anthropic, Build live dashboards and animate explainers with Claude (October 8, 2026: Docs, Slides, and Design out of beta; on by default for Enterprise October 15; Dashboards and Motion off by default for Enterprise; external sharing only if the admin allows it)
- Anthropic, Your apps can now open in Claude, and Claude can work in them (November 10, 2026)
- Microsoft, Configure Microsoft Copilot settings in the Microsoft 365 admin center (Microsoft Learn)
- Microsoft, Assign user licenses and manage access for Copilot Studio (Microsoft Learn)
- Microsoft, Get started deploying Copilot and agents (Microsoft Copilot Control System playbook: pilot groups, license assignment through Entra ID groups)
- Microsoft, October 2026 announcements (Partner Center: "(add-on)" names from October 16, new offers without a base subscription from December 1)
- OpenAI, Admin controls, security, and compliance for plugins and apps (OpenAI Help Center: Business and Enterprise defaults, role-based access, plugins and apps)
- OpenAI, ChatGPT Enterprise admin quickstart (OpenAI Help Center)
- McKinsey, The state of AI in 2025: Agents, innovation, and transformation (November 5, 2025)