A new era of AI assistance is coming to Microsoft Teams. According to the Microsoft 365 public roadmap, the feature for “Interactive Agents for Teams Meetings and Calls” is scheduled to begin its rollout in September 2026. This headline addition will bring autonomous Copilot agents directly into Teams meetings and 1-on-1 calls, complete with group chat and meeting chat integration. For IT professionals and decision makers, the clock is already ticking. The intervening months demand a proactive approach to governance, a clear understanding of session memory implications, and a concrete plan for enablement.

Interactive Agents: More Than a Chatbot

Microsoft has been steadily expanding the capabilities of Copilot across Microsoft 365, and the next frontier is agentic AI. These interactive agents, built on Microsoft Copilot Studio, will not merely respond to text prompts in a sidebar. They will join meetings as full participants, process spoken conversations, access meeting transcriptions, and interact with users through the meeting chat. The roadmap excerpt indicates that the feature will be available for both group meetings (two or more participants) and 1-on-1 calls, significantly broadening the scope of autonomous AI interaction in real-time collaboration.

Early documentation and demos from Microsoft have shown agents performing tasks such as retrieving information from connected knowledge bases, summarizing discussion points, tracking action items, and even facilitating Q&A sessions within the meeting context. Unlike the existing Copilot in Teams meeting recap, which provides a post-meeting summary, these agents will operate in the moment—listening, reasoning, and responding as the meeting unfolds. For organizations that have invested in custom agents via Copilot Studio, the September 2026 rollout will allow those agents to be deployed directly into live meetings, acting on behalf of specific departments or business processes.

Agent Architecture and Permissions

Interactive agents in meetings will inherit the permission model of Copilot Studio and Microsoft 365. An agent’s access to meeting data—such as transcription, chat messages, and shared files—will be governed by the organizer’s tenant settings and the specific agent configuration. Agents will authenticate via the Microsoft Entra ID of the user who added them to the meeting or via a service principal, depending on how IT admins have configured the service. Crucially, agents will not have unrestricted access to all meeting data by default; they will operate within the same compliance boundaries as other meeting participants, subject to policies for data loss prevention, information barriers, and retention.

Meeting Governance: Defining the Rules of Engagement

The introduction of interactive agents into live meetings raises a host of governance questions. Unlike a passive note-taker, an active agent can speak, post in chat, and even modify meeting artifacts if granted permission. IT admins will need to establish clear policies before the September 2026 launch to prevent sprawl, protect sensitive data, and maintain meeting integrity.

Admin Controls in Teams Admin Center

At the time of rollout, Microsoft is expected to surface agent governance controls within the Teams Admin Center and the Copilot Studio admin portal. Current controls for Microsoft 365 Copilot in meetings (which will likely be extended) include the ability to allow or block AI features at the tenant level, for specific users or groups. For interactive agents, additional granularity will be required:

  • Agent availability policies: Admins can choose which agents (by app ID) are permitted to join meetings, and under what conditions.
  • Meeting options: Organizers and co-organizers will be able to toggle whether agents can join a specific meeting, similar to the existing “Allow attendance report” or “Allow live captions” toggles. This may appear as a new “Allow interactive agents” setting.
  • Permission scopes: Admins will be able to restrict agent actions—for example, allowing agents to read chat but not to post, or allowing voice synthesis only with explicit organizer approval.
  • Data handling during meetings: Controls to specify whether agent data processing is performed entirely within the Microsoft 365 trust boundary (in-region processing) or if it may involve external AI services (as some Copilot Studio integrations do).

Organizations in regulated industries—financial services, healthcare, government—must assess how agent participation may affect records management, eDiscovery, and client-attorney privilege. If an agent generates content or decisions during a meeting, that content becomes part of the meeting record. IT and legal teams should plan to:

  • Update eDiscovery workflows to include agent-generated content from meeting transcripts and chat logs.
  • Review retention policies to confirm that agent data is covered by standard lifecycle management rules.
  • Evaluate agent access to confidential information under data classification and information protection frameworks.
  • Conduct privacy impact assessments for EU data subjects under GDPR, given the real-time processing of biometric voice data.

Microsoft typically provides documentation and compliance alignments ahead of major feature releases. IT departments should expect detailed technical documentation at least six months prior to the September 2026 rollout, allowing time for internal audits and policy creation.

Session Memory: Between This Meeting and the Next

One of the most intriguing—and potentially disruptive—capabilities of interactive agents is session memory. In a single meeting, agents will remember earlier parts of the conversation to maintain context. But what happens after the meeting ends? Will agents retain memory across multiple sessions, learning from past interactions with a specific user or group?

Ephemeral vs. Persistent Memory

Microsoft has not yet disclosed the full memory model for Teams interactive agents, but existing Copilot features offer clues. Current meeting recaps use the meeting transcript as source material and do not retain per-user conversation memory across separate meetings. However, Copilot Studio agents can be configured with persistent knowledge—connected data sources like SharePoint, Dataverse, or custom APIs—that do carry over across sessions. For interactive agents, the likely architecture is:

  • Ephemeral session memory: Within a single meeting, the agent maintains a context window covering recent turns of the conversation. This context is used to answer follow-up questions, refer to previous statements, and ensure coherent dialogue. The context window will have a token limit, and older parts of the meeting may fall out of scope as the meeting progresses.
  • Persistent knowledge: Data sources connected through Copilot Studio are always available, and the agent can reference them across meetings. For example, a sales agent can pull CRM data into any meeting where it is invoked, regardless of session boundaries.
  • Cross-meeting memory (per user): Microsoft may introduce optional memory that allows an agent to remember user preferences or past interactions across meetings. If implemented, this would require explicit user consent and would fall under the user’s own Microsoft 365 storage (e.g., in the Semantic Index for Copilot). IT admins will likely have controls to enable or disable such persistent memory at the tenant or user level.

Privacy and Data Minimization

Session memory raises immediate privacy concerns. Meeting organizers and participants must be clearly informed when an agent is present and what it remembers. Microsoft will need to provide transparency features, such as:

  • A visual indicator in the Teams UI showing that an interactive agent is active.
  • A summary at the end of the meeting of what data the agent stored or learned.
  • A means for users to delete agent memory associated with them, either per meeting or globally.

For IT, the key preparatory step is to align agent memory policies with existing data governance frameworks. If an agent stores information about users, that storage must be accounted for in data retention schedules, and users must be able to exercise their data subject rights under regulations like GDPR and CCPA.

IT Preparation: A Three-Phase Roadmap to Readiness

The September 2026 date gives organizations roughly 18 months to prepare, assuming typical Microsoft roadmap cadence. A phased approach will ensure a smooth deployment and minimize disruption.

Phase 1: Discovery and Assessment (Now – Q3 2025)

  • Monitor roadmap updates: The exact feature ID for “Interactive Agents for Teams Meetings and Calls” is not yet public, but it will appear in the Microsoft 365 roadmap. Subscribe to the Microsoft 365 message center for related announcements.
  • Identify use cases: Work with business units to pinpoint meetings where an interactive agent would deliver value—status meetings, customer calls, training sessions, etc. Document both the expected benefits and the potential risks for each scenario.
  • Inventory existing Copilot Studio investments: Determine which departments already build custom agents and how those agents might be extended to meetings.
  • Review licensing: Current Microsoft 365 Copilot licensing ($30 per user/month) does not include the full scope of Copilot Studio for building custom agents. Interactive agents in meetings may require an additional Copilot Studio license (around $200 per month for 25,000 messages). Audit your current license position and budget accordingly.

Phase 2: Governance Design and Testing (Q4 2025 – Q2 2026)

  • Define governance policies: Draft a policy document specifying who can add agents to meetings, which agent actions are allowed, and how agent data is handled. Circulate for legal and compliance review.
  • Configure pilot settings: As Microsoft releases tenant-level controls (likely in private preview or early public preview in mid-2026), configure a pilot group. Start with a small set of trusted users and a single agent scenario.
  • Test agent behavior: Evaluate how agents interact with live meetings—do they interrupt? How accurate are their responses? How do participants react? Gather qualitative feedback from pilot users.
  • Validate compliance: Run a test eDiscovery case including agent-generated content. Confirm that retention labels apply correctly to agent data in meeting transcripts. Verify that data residency requirements are met.

Phase 3: User Enablement and Rollout (Q3 2026)

  • Communicate ahead of launch: Notify users about the availability of interactive agents, the policies governing their use, and the benefits they bring. Provide training materials or short videos demonstrating proper interaction with agents.
  • Configure meeting templates: Use Teams meeting templates to set appropriate defaults for agent availability. For example, a “Confidential Board Meeting” template could disallow agents by default, while a “Project Standup” template could allow them.
  • Set up monitoring and feedback channels: Establish a process for users to report unexpected agent behavior or privacy concerns. Monitor usage patterns via the Teams Admin Center to identify proliferation of unsanctioned agents.
  • Iterate: After initial rollout, gather data on agent usage and iterate on your governance policies. Microsoft may release additional controls or memory features post-launch; stay engaged with the roadmap.

The User Experience: What Meeting Participants Will See

When an organizer adds an interactive agent to a meeting, the agent will appear in the participant list with an identifying badge—likely similar to the existing Copilot icon. In a meeting with an agent present, all participants will see a persistent visual cue, ensuring that no one is unaware they are interacting with an AI. Agents may contribute via the meeting chat or, if voice synthesis is enabled, speak during the call. Organizers will have the ability to mute or remove the agent at any time, just like any participant.

Users will interact with the agent through standard chat messages, possibly using a dedicated command prefix or natural language. For example, a user might type “@Agent, what were the action items from the last project update?” and the agent would query its memory and respond in chat. In voice-enabled scenarios, participants could ask questions aloud, and the agent would respond via synthesized speech—though Microsoft will likely limit voice participation to avoid conversational chaos.

The Broader AI Governance Picture

Interactive agents in Teams are just one piece of Microsoft’s agentic AI push. Copilot Studio agents are also appearing in Outlook, SharePoint, and other Microsoft 365 surfaces. The governance patterns established for Teams meetings will set precedents for how agents are managed across the entire productivity suite. Organizations that invest early in a cross-platform agent governance framework will be better positioned to handle these expansions.

Key principles that should guide any framework include:

  • Least privilege: Agents should only have the permissions necessary for their defined role.
  • Data minimization: Memory and data collection should be limited to what is required for the agent to function.
  • Transparency: Users must always know when an agent is present and what it can do.
  • Auditability: All agent actions should be logged and retrievable for compliance purposes.

Microsoft’s own Responsible AI principles and the Copilot Trust Center will provide foundational guidance, but each organization must tailor controls to its risk appetite.

Looking Ahead

The September 2026 rollout is a target date, and Microsoft roadmaps can shift. However, the trajectory is clear: AI agents are becoming full-fledged meeting participants. For IT leaders, the message is equally clear—start your governance planning now. The difference between a chaotic agent rollout and a strategic AI implementation will be measured in the policies, testing, and user education completed before the first agent says “hello” in a Teams meeting.

By proactively addressing meeting governance, session memory, and IT readiness, organizations can harness the productivity gains of interactive agents while maintaining the trust and security of their collaboration environments.