AI agent sprawl is the new Teams sprawl

If you spent 2021 cleaning up Microsoft Teams that nobody owned, the next cleanup is already forming in your tenant. Agents are being created the way Teams were in 2020: open by default, no owner required, and no record of what they can reach. AI agent sprawl is the uncontrolled growth of AI agents across your Microsoft 365 tenant, multiplying faster than anyone can track who built them or what data they touch, and you've seen how this ends.

The pattern is familiar because the defaults are identical. What's different this time is what an ungoverned object can do once it's loose in your environment.

What AI agent sprawl is, and why it's happening now

AI agents now come from several places at once in Microsoft 365: Copilot Studio, SharePoint, and Agent Builder. An agent is just an application that uses AI to do a task, but the number of front doors for creating one is what turns it into sprawl. Since Microsoft Agent 365 reached general availability on May 1, 2026, building an agent is within reach of anyone with a license, and adoption has moved from a handful of pilots to something happening in every department. If you want a refresher on the types and what they do, start with what Copilot agents actually are.

SharePoint agents are the fastest to multiply. Any member with create rights on a site can make one, they're stored as .agent files on the site itself, and they inherit that site's sharing scope. An agent can be created, shared, and forgotten without IT ever touching it. Multiply that across every site in a mid-size tenant and the roster gets long fast.

This isn't a Copilot-only problem, either. The same conditions apply regardless of which AI you use, whether that's a Microsoft 365 Copilot agent, a third-party model, or something your developers wired up themselves.

Why agent sprawl is the new Teams sprawl

Rewind to 2020. Microsoft 365 shipped with workspace creation open by default. Anyone could create a Team, there were no naming conventions, no owner requirements, and IT had no visibility into what was being built until after it was created. By 2022, most tenants had hundreds of workspaces nobody owned and a cleanup project nobody had planned for. Some organizations are still working through it. For the full history, see what Teams sprawl is and how it happens.

That cleanup is a big part of why Orchestry exists. The same open-creation defaults that produced Teams sprawl are now producing agents, and the trajectory looks the same: easy creation, no accountability, and a bill that comes due later.

The difference is that IT has run this play before. You already know what "later" looks like, which means you can act before the roster gets unmanageable rather than after it does.

Why ungoverned agents raise the stakes

A stray Teams channel is a container. The worst case is data you can't find and an owner you can't reach. An agent is different, because it acts on data: it connects to the sites and files its builder wired in, and it can surface whatever those connections allow.

That inheritance is the sharp edge. A SharePoint agent is stored on its site and takes on that site's sharing, so it can surface whatever the site already exposes. Agents respect each user's permissions, which means the real risk is content that was already overshared: the agent just makes it far easier to pull up. One overshared site is all it takes to turn a single agent into a data-exposure problem.

So an agent built by someone in finance, pointed at sensitive files, shared more broadly than intended, and owned by someone who has since left, is a live exposure sitting quietly in your tenant. The same accumulation problem that left tenants full of ownerless Teams is already showing up with agents: experimental builds that outlived their purpose, creators who changed roles, and no process for what happens to an agent when the person who made it leaves.

Agents also drift. One built today with narrow access picks up a new connector six months from now, the scope widens, and nobody is notified. Microsoft's own guidance now treats agents as a governance problem in their own right, a shift we covered after Ignite 2025.

There's a cost to all of this, too. Every ungoverned agent adds consumption charges and cleanup debt, landing on top of the Microsoft 365 licensing changes teams are already absorbing this year. It's a bill nobody planned for, growing quietly next to the roster you can't see.

Shadow AI agents and the tenant visibility gap

Before you can govern any of this, you have to answer a basic question: how many agents exist, who owns them, and what are they connected to? Most organizations can't. To assemble the picture today, an admin moves across five separate Microsoft portals, the Microsoft 365 admin center, SharePoint admin, Power Platform admin, Purview, and Defender, and even then the result is a set of object IDs rather than owner names.

Microsoft says it plainly in its Cloud Adoption Framework: "you can't govern agents you don't know exist." The unmanaged ones have a name now, shadow AI agents, and Microsoft has started surfacing them through a Shadow AI page in the Microsoft 365 admin center, currently in preview. The free admin view can tell you an agent exists and flag a high-severity count, but a count isn't a governance decision. The visibility gap is the starting problem, not the frameworks that come after it.

What Microsoft Agent 365 does, and where the gap remains

Microsoft's answer is Agent 365, a control plane to observe, govern, and secure agents. It gives admins a registry and a central place to see what's running, which confirms the point of this whole piece: visibility comes first.

Seeing the agents is step one, not the finish line. Knowing an agent exists isn't the same as knowing whether it should, and that's a governance question, not an inventory one.

AI agent governance starts with the playbook you already ran

Agentic AI governance isn't a new discipline. It's the model you already built for Teams, applied one layer up. With Teams, IT eventually moved from open creation to governed self-service: a request fires a template, naming and policies attach at creation, ownership is required before anything goes live, and a lifecycle review is on the calendar from day one. For the mechanics, see how governed provisioning replaces open creation.

That's the model Orchestry runs today for workspaces. Orchestry's provisioning templates carry governance in at creation, so naming, sensitivity labels, sharing rules, and a named owner are set the moment a workspace is born rather than bolted on later. Its delegated lifecycle workflows route archive-and-retire decisions to the workspace owner instead of IT, which is how Orchestry customers clean up 75% of unused Teams within the first 14 days of turning it on.

The reason that matters for agents is scale. Based on Orchestry data, 67% of workspaces show no activity in the trailing 90 days when organizations first connect to the platform. Sprawl hides in plain sight, and it takes a recurring review cadence, not a one-time audit, to keep it in check. That's the muscle memory agent governance will need.

Applying that model to agents is where Orchestry is headed next. Orchestry is building AI and agent governance to bring agents into the same operating model it already uses for workspaces, so the roster you can finally see becomes a roster you can actually manage. It's coming later this year.

The question isn't whether you can govern agents. You've governed uncontrolled creation before. The question is whether you start before the sprawl builds, or after.

AI agent sprawl: frequently asked questions

What is AI agent sprawl?

AI agent sprawl is the uncontrolled growth of AI agents across a Microsoft 365 tenant, created faster than anyone can track who owns them or what data they can reach. It happens because agent creation is open by default, so agents accumulate across Copilot Studio, SharePoint, and Agent Builder without ownership or review.

How is agent sprawl different from Microsoft Teams sprawl?

The creation pattern is the same, but the risk is higher. A sprawled Teams workspace is a container of content, while a sprawled agent actively connects to data and can surface whatever its permissions allow. That makes an ownerless or over-shared agent a live exposure, not just clutter.

How can I see what AI agents already exist in my Microsoft 365 tenant?

Today it means checking several Microsoft admin portals, and Microsoft's Agent 365 registry consolidates that view for organizations that adopt it. The first move that matters is building a single inventory of every agent with a named owner, because you can't govern what you can't see.

Get ahead of agent sprawl

The Teams sprawl of 2020 taught IT what open-by-default creation costs. Agents are running the same play, and you don't have to wait for the cleanup to start.

Part 1 of our Unseen Agents webinar series, the agents already in your tenant, makes the problem real before it becomes a cleanup project. Michal Pisarek and David Francoeur host on July 20.