Microsoft’s roadmap for Teams Interactive Agents now undercuts its own September 2026 rollout target with a stark admission: the company “cannot continue rolling this out at this time.” The conflicting status, first spotted in Microsoft 365 Roadmap item 164871, leaves IT administrators with a planning marker — but no reliable deployment commitment.
What Interactive Agents Promise—and What the Roadmap Actually Says
Interactive Agents are designed to bring AI directly into Teams meetings and 1:1 calls. According to the roadmap entry, participants will be able to interact with these agents privately or as a group, use zero-state prompts, and access conversation history. On paper, it’s a leap forward for meeting intelligence.
The roadmap currently shows September 2026 as the rollout start date. But that date is immediately undermined by the entry’s own status language. The same text that points to September also states that Microsoft is “not able to continue rolling this out at this time.” That contradiction — a scheduled start alongside an admission the rollout can’t proceed — signals that September is, at best, a placeholder. It is not a committed release window.
How We Got Here: The Uncertain Road to AI in Meetings
Microsoft has been layering AI into Teams for years, from transcription and live captions to Copilot integration. Interactive Agents represent the next phase: purpose-built AI participants that join calls just like a person. The feature first appeared on the roadmap months ago with a distant target date, a common practice for emerging capabilities.
However, the status update throws that timeline into doubt. It’s possible that the roadmap item was published prematurely, or that internal testing hit a snag. Either way, administrators should not treat September 2026 as a change window in their production environments. The entry lacks standard release details — no indication of which channels or tenants will see it first, no prerequisites listed, and no confirmation that the feature is even ready for preview.
What This Means for IT Administrators
Organizations that lock in training schedules, user communications, or change board approvals based on the September date are taking a gamble. The responsible approach is to move that marker to a risk register and a roadmap watchlist. It belongs there, not in a production deployment plan.
That doesn’t mean IT teams should wait. Governance cannot be retrofitted after agents start appearing in meetings. Now is the time to build the policies and controls that will apply whenever Interactive Agents finally arrive — whether that’s in September, later, or in a limited preview ring.
Step 1: Understand the Two Classes of Agents
Not all agents are equal, and a one-size-fits-all policy will break. Microsoft draws a critical line between agents built on Copilot and custom agents. Only Copilot-based agents retain conversational context across turns within the same session; custom agents do not yet support sessions.
This distinction matters for testing and disclosure. A Copilot agent’s behavior may shift mid-meeting based on what was said earlier, while a custom agent responds without that continuity. Test plans must account for agent type, and any policy that assumes persistent session memory across all agents is inaccurate until Microsoft updates custom agent capabilities.
Step 2: Use Existing Admin Controls to Restrict Access
The Teams admin center already provides a starting point. Under Teams apps > Built-in Teams agents, administrators can set availability to everyone, selected users or groups, or no one. This page currently manages agents like Channel Agent and Facilitator, and it’s the logical home for Interactive Agents control.
But a warning: Microsoft hasn’t confirmed exactly how Interactive Agents will appear in the admin experience. The roadmap doesn’t specify which page or setting to use. So while the built-in agents page is the best current bet, IT must verify what’s actually exposed in their tenant when the feature does show up. Relying on the page without checking could leave gaps.
Org-wide app settings and app-centric management offer additional levers. Starting April 2025, tenants were automatically migrated to app-centric management, which lets admins control per-user and per-app availability. These tools can enforce the initial narrow deployment that every pilot demands.
Step 3: Define Meeting Policies Before Agents Join
The toughest question isn’t whether an agent can join a meeting. It’s what data the agent can receive, retain, produce, and potentially pass to another workflow. Organizers should never make these decisions on the fly.
A simple decision matrix, adapted from governance best practices, can anchor those rules. Use it to classify every meeting scenario before granting any agent access:
| Meeting Condition | Baseline Decision Before Pilot Use |
|---|---|
| Internal routine meeting | Permit only for approved test group; participant notice required; named organizer responsible. |
| 1:1 call | Require explicit notice; private interactions and session history may be less visible to the other participant. |
| External participants present | Require organizer review and a defined method for handling objections before agent interaction begins. |
| Restricted or sensitive material | Block by default until security, privacy, and compliance owners approve a documented exception. |
| Recording or transcription enabled | Confirm separately whether agent use, recording, transcription, and resulting artifacts are each permitted. |
| Agent ownership or behavior cannot be verified | Do not admit or enable until owner, purpose, and approved data use are established. |
This matrix is an organizational starting point — not a substitute for legal or contractual review. Consent and notice requirements vary by jurisdiction and workforce agreements.
Step 4: Set Up Your Pilot Governance Checklist
A controlled pilot is the only safe way to introduce any agent. But ‘controlled’ means more than limiting the user count. It means documented ownership, defined boundaries, and a working kill switch tested before the first meeting.
Before anyone enables an Interactive Agent in even a test call, complete these seven actions:
- Assign an accountable service owner. One named Teams or Microsoft 365 service owner must hold authority over availability, rollout rings, exceptions, suspension, and communications.
- Create a restricted access group. Limit the first deployment population to designated testers — not the entire tenant. Include representatives from admin, security, compliance, privacy, support, and business units.
- Approve meeting scenarios explicitly. Decide which meetings may include agents: internal, external, 1:1, those with restricted data, or sessions with recording/transcription.
- Define participant-notice requirements. Specify who announces the agent, when, what the notice explains, and what happens if a participant objects or cannot consent under your policies.
- Separate agent use from recording decisions. Permitting an agent does not automatically authorize recording, transcription, or retention. Each capability needs its own explicit approval.
- Document the shutdown path. Know how to change an agent’s availability to “no one,” remove the pilot group, suspend an individual agent, and notify organizers. Prove it works before the first pilot session.
- Set an evidence standard. Retain approvals, policy configurations, test results, exception records, and the final go/no-go decision according to your organization’s retention rules.
If Microsoft later changes the roadmap date or the session memory model, that evidence trail shows which assumptions must be retested.
Step 5: Don’t Confuse External Bot Approval with Agent Control
Microsoft already has a lobby-based guardrail for external bots in meetings. By default, organizers must approve any detected external bot before it can enter. That’s useful for third‑party transcription services, but it’s not a universal agent switch.
The lobby policy catches external bots. Interactive Agents, especially built‑in or Copilot‑based ones, may not pass through that same checkpoint. Assuming otherwise could leave internal agents unscrutinized.
Keep the lobby approval for external bots. For sanctioned Teams agents, create a separate allowlist and access group. Train organizers to reject unfamiliar entries, even those with benign‑sounding labels. The person running the meeting shouldn’t become the sole security gatekeeper without support.
Step 6: Ensure You Can Shut It Down Fast
A kill switch that hasn’t been tested is just a checkbox in a plan. Before the first pilot meeting, Teams administrators should prove they can change a built‑in agent’s availability to “no one” and then revert it. Know who can authorize that action outside normal change windows, and make sure the process works within seconds, not hours.
Post‑rollout reviews are equally important. Audit pilot membership regularly to ensure it hasn’t grown informally. Each enabled agent should still have an active owner. Meeting scenarios must stay within the approved scope. Exceptions need expiration dates and designated approvers; a permanent exception created to unblock one meeting can become a shadow policy.
The Bottom Line: Prepare for September, But Don’t Bet On It
The September 2026 roadmap marker is useful as a near‑term deadline for governance preparation. It is not yet reliable enough to anchor production deployment. Organizations that build their controls now will be ready whenever Microsoft clears up the confusion. Those that treat the date as a firm release window risk scrambling to retrofit policies after agents are already live.
Outlook: What to Watch For
Microsoft still owes administrators several clarifications: whether Roadmap ID 164871 has actually resumed, which tenants and release channels will get it first, what prerequisites apply, and how the admin controls will map to the Interactive Agents experience. Watch for updates to the roadmap, announcements in the Message Center, and changes in the Built‑in Teams agents page in the admin center. The organizations best prepared for September are the ones that have their governance in place — and are equally ready for September to pass without a production rollout.