5.5 Sharing an Agent with Team Members

Key Takeaways

  • Copilot Agents support multiple deployment tiers, ranging from private personal drafts and direct link sharing to Microsoft Teams channel embedding and tenant-wide organizational catalog publishing.
  • Agents strictly enforce User Identity Passthrough: all data retrieval executes under the runtime caller's Microsoft Entra ID credentials, meaning agents cannot bypass SharePoint ACLs or Microsoft Purview sensitivity labels.
  • Organizational publishing of custom declarative agents requires IT administrator review and approval within the Microsoft 365 Admin Center before becoming discoverable tenant-wide.
  • Maintaining agent health requires regular lifecycle management, including synchronizing modified knowledge repositories, auditing response accuracy, and decommissioning obsolete assistants.
  • IT administrators govern agent creation, installation, and data access policies through the Microsoft 365 Admin Center, Copilot Studio Admin Center, and Microsoft Purview Data Loss Prevention (DLP) controls.
Last updated: August 2026

5.5 Sharing an Agent with Team Members

Quick Answer: Deploying and governing Copilot Agents follows a structured lifecycle across four distribution tiers: Personal Draft, Direct Link Sharing, Team/Channel Embedding, and Organizational Catalog Publishing (which requires IT administrator approval). Crucially, all Copilot Agents operate under User Identity Passthrough—an agent always executes data retrieval using the runtime caller's Microsoft Entra ID identity token, never the creator's elevated credentials. As a result, agents strictly honor SharePoint Access Control Lists (ACLs) and Microsoft Purview sensitivity labels. Enterprise governance is managed centrally through the Microsoft 365 Admin Center, Microsoft Copilot Studio Admin Center, and Purview audit logging.

Building a high-performing declarative agent is only half the equation in enterprise generative AI adoption. To drive tangible business transformation, organizations must deploy agents effectively across teams while maintaining rigorous data security, compliance governance, and lifecycle hygiene. A failure in governance can lead to sprawl, reliance on stale knowledge repositories, or confusion surrounding data access permissions.


1. Agent Distribution and Deployment Models

Microsoft 365 Copilot provides flexible deployment models tailored to different collaboration scopes—from individual productivity tools to mission-critical enterprise services.

+-------------------------------------------------------------------------+
|                   AGENT DISTRIBUTION & DEPLOYMENT TIERS                 |
+-------------------------------------------------------------------------+
|  [TIER 1: PERSONAL USE]          -> Private author draft; testing &     |
|                                     personal productivity.              |
|                                                                         |
|  [TIER 2: DIRECT LINK SHARING]   -> Share secure link with colleagues   |
|                                     or specific security groups.        |
|                                                                         |
|  [TIER 3: TEAMS & CHANNELS]      -> Embed/pin agent in Teams channels,  |
|                                     group chats, & SharePoint pages.    |
|                                                                         |
|  [TIER 4: ORGANIZATIONAL STORE]  -> Submit for IT Admin review & publish|
|                                     to tenant-wide agent catalog.       |
+-------------------------------------------------------------------------+

The 4 Deployment Tiers Explained

Tier 1: Personal Use (Private Draft)

  • Scope: Accessible only to the agent's creator.
  • Lifecycle State: Draft / Prototyping.
  • Use Case: Iterating on system instructions, testing knowledge grounding, validating capability toggles, and refining prompt starters before broader sharing.

Tier 2: Direct Sharing via Link (Peer-to-Peer)

  • Scope: Any colleague or security group who receives the sharing link generated by the author.
  • Experience: When a recipient clicks the link, the agent is automatically added to their personal Microsoft 365 Copilot Chat right-rail agent menu.
  • Use Case: Small project teams, specialized working groups, or pilot user acceptance testing (UAT).

Tier 3: Teams Channels, Group Chats & SharePoint Pages

  • Scope: All members of a designated Microsoft Teams team, channel, group chat, or SharePoint site.
  • Experience: Team members can @mention the agent directly inside channel threads or group chats to answer collaborative questions in real time. The agent can also be embedded as a web part on SharePoint communication sites.
  • Use Case: Project war rooms, regional sales channels, sprint planning groups, and departmental intranet portals.

Tier 4: Organizational Publishing (Enterprise Agent Store)

  • Scope: Tenant-wide availability for all licensed Microsoft 365 Copilot users in the organization.
  • Governance Requirement: Requires formal submission by the creator and IT Administrator approval in the Microsoft 365 Admin Center.
  • Experience: The agent appears in the central Copilot Agent Store under the "Built for your org" category.
  • Use Case: Universal enterprise services such as Corporate IT Helpdesk, Global HR Benefits, Travel & Expense Compliance, and Brand Asset Guidelines.

Deployment Tier Comparison Matrix

Deployment TierDistribution MechanismAdmin Approval Required?Target AudienceIdeal Business Scenario
Personal UsePrivate author workspaceNoIndividual creatorPersonal productivity, prompt testing, RAG validation
Direct LinkHTTPS shareable URLNoSpecific peers or security groupsPilot testing, small project teams, confidential task forces
Teams & ChannelsChannel app installation / @mentionNo (if tenant policy permits)Team / channel membersCollaborative triage, sprint retrospectives, customer account channels
Org CatalogCentral Copilot Agent StoreYes (Mandatory IT Admin approval)Entire tenant or targeted M365 security groupsGlobal enterprise services (HR, IT Helpdesk, Corporate Travel)

2. Access Control, Security & User Identity Passthrough

A critical area of testing on the AB-730 exam centers on security architecture and access control. A widespread misconception among business users is that creating an agent grants its users elevated access to the creator's files. In Microsoft 365 Copilot, this is strictly impossible due to User Identity Passthrough.

┌─────────────────────────────────────────────────────────────────────────────┐
│                     USER IDENTITY PASSTHROUGH IN ACTION                     │
│                                                                             │
│  [Creator User A] ──► Configures Agent with Knowledge:                      │
│                       SharePoint Site: /Confidential_Executive_M&A/         │
│                                                                             │
│  [User B (No ACLs)] ──► Submits Prompt to Shared Agent                      │
│                             │                                               │
│                             ▼                                               │
│  [Copilot Orchestrator] ──► Evaluates Entra ID Token for USER B             │
│                             │                                               │
│                             ▼                                               │
│  [Microsoft Graph] ───────► Checks SharePoint ACLs for USER B               │
│                             │                                               │
│                             ▼                                               │
│  [Access Denied] ─────────► User B lacks read permissions                   │
│                             │                                               │
│                             ▼                                               │
│  [Grounded Output] ───────► "I cannot find any relevant documents in the   │
│                             knowledge sources accessible to you."           │
└─────────────────────────────────────────────────────────────────────────────┘

The Golden Rules of Copilot Agent Security

  1. User Identity Passthrough: When a user queries a declarative agent, the agent never executes queries using the author's credentials. Instead, all Microsoft Graph searches and document retrievals execute under the runtime caller's Microsoft Entra ID identity token.
  2. Strict Access Control List (ACL) Enforcement: If an agent's configuration includes a SharePoint document library containing 100 files, but User B only has read access to 10 of those files, the agent can only retrieve and ground answers on those 10 files when User B is chatting. For the other 90 files, the agent behaves as if they do not exist.
  3. No Privilege Escalation: An agent cannot be used as a backdoor or proxy to view restricted payroll data, executive compensation sheets, or confidential project files.
  4. Microsoft Purview Sensitivity Label Inheritance: If a source document is protected by a Microsoft Purview Information Protection (MPIP) sensitivity label with encryption (e.g., "Confidential / C-Suite Only"), an employee lacking rights to open that labeled file will receive zero content from that file when interacting with the agent. Furthermore, responses generated from labeled documents respect tenant DLP and sensitivity classification rules.

Exam Warning: If an exam question describes a scenario where an executive creates an agent pointing to a private executive SharePoint site and shares the agent link with an intern, the intern CANNOT see or summarize the executive documents. The agent will inform the intern that no relevant documents were found. This is due to identity passthrough and security trimming.


3. Agent Lifecycle Management: Monitoring Health, Updating Knowledge & Decommissioning

Deploying an agent is not a one-time project; it requires continuous lifecycle management to ensure operational accuracy, compliance, and user satisfaction.

+-------------------------------------------------------------------------+
|                       AGENT LIFECYCLE PHASES                            |
+-------------------------------------------------------------------------+
|  [1. DESIGN & AUTHOR]    ──► Draft metaprompt, scope knowledge, toggle   |
|                              capabilities, & configure starters.        |
|                                                                         |
|  [2. PILOT & VALIDATE]   ──► Share via link with test cohort; audit     |
|                              citations & evaluate fallback responses.   |
|                                                                         |
|  [3. DEPLOY & ADOPT]     ──► Publish to Org Store or pin in Teams;      |
|                              track active usage & user feedback.        |
|                                                                         |
|  [4. MAINTAIN & SYNC]    ──► Refresh knowledge files, update URLs, &    |
|                              refine prompt guardrails quarterly.        |
|                                                                         |
|  [5. DECOMMISSION]       ──► Unpublish from store, revoke sharing links,|
|                              & archive to prevent stale guidance.       |
+-------------------------------------------------------------------------+

Operational Maintenance Tasks

  • Knowledge Synchronization & Staleness Management: As corporate policies evolve, knowledge repositories must be maintained. When files in an attached SharePoint site are updated or replaced, the Semantic Index automatically updates its vector embeddings. However, if folder structures change or files are renamed, agent owners must verify that knowledge URLs remain valid and remove deprecated reference documents.
  • Performance Monitoring & Feedback Loops: Agent creators and department leads should review qualitative user feedback (thumbs-up / thumbs-down ratings in Copilot Chat) and analyze failure patterns where the agent triggers fallback responses (e.g., "I do not have access to that information"). These signals identify knowledge gaps that require adding new documentation or adjusting system instructions.
  • Decommissioning Obsolete Agents: When a project completes or an annual policy is superseded (e.g., "Fiscal 2024 Benefits Agent"), the agent owner or IT admin must unpublish the agent from the organizational catalog, revoke sharing links, and archive the agent definition in Copilot Studio. This prevents employees from making business decisions based on obsolete AI guidance.

4. IT Administrative Governance & Tenant Controls

Enterprise IT administrators oversee the entire Copilot Agent ecosystem through the Microsoft 365 Admin Center, Microsoft Copilot Studio Admin Center, and Microsoft Purview.

+-------------------------------------------------------------------------+
|                 ENTERPRISE IT ADMIN GOVERNANCE STACK                    |
+-------------------------------------------------------------------------+
|  [M365 ADMIN CENTER]             [COPILOT STUDIO ADMIN]                 |
|  • Manage Integrated Apps /      • Manage Power Platform environments.  |
|    Agent Management portal.      • Configure Data Loss Prevention (DLP) |
|  • Approve/reject Org Store        connector policies.                  |
|    publishing submissions.       • Control user creation rights.        |
|  • Assign agent deployment       • Monitor tenant capacity & compute.   |
|    policies per security group.  • Review cross-environment telemetry.  |
|                                                                         |
|  [MICROSOFT PURVIEW]                                                    |
|  • Audit log search: record every prompt, response, & agent invocation. |
|  • Enforce sensitivity labels & Information Protection encryption.      |
|  • eDiscovery & legal hold across all Copilot Agent conversations.      |
+-------------------------------------------------------------------------+

Key Administrative Controls

  1. Integrated Apps & Agent Management (M365 Admin Center):
    • Agent Creation Policies: Administrators can control who in the organization is permitted to build and share custom declarative agents (e.g., enabling creation for all users, restricting creation to specific security groups, or disabling it globally).
    • Reviewing Organizational Submissions: When a creator submits an agent to the organization catalog, IT admins inspect the agent's manifest, verify its attached knowledge sources, ensure compliance with data governance policies, and approve or reject publication.
    • Centralized App Deployment: Admins can automatically deploy and pin critical agents (e.g., Corporate IT Helpdesk) directly to the Copilot right rail or Teams navigation for all licensed employees or specific departments.
  2. Environment & DLP Controls (Copilot Studio Admin Center):
    • Environment Routing: Directing agent development to dedicated sandbox or production environments within the Power Platform tenant.
    • Data Loss Prevention (DLP) Policies: Restricting what connectors an agent can utilize (e.g., blocking unapproved external HTTP webhooks or social media APIs while permitting internal SharePoint and Microsoft Graph connectors).
  3. Audit Logging & Compliance (Microsoft Purview):
    • Every interaction with a Copilot Agent generates a structured event in the Microsoft Purview Unified Audit Log. Compliance officers can review prompt text, retrieved file metadata, and timestamps for regulatory audits and eDiscovery investigations.

5. Realistic Business Scenario: Enterprise Rollout of Contoso's Global Benefits Agent

The Business Requirement

Contoso Health Systems operates hospital networks across North America and the United Kingdom. HR designs a Global Benefits & Wellness Agent to help 25,000 employees navigate health insurance, retirement plans, parental leave, and wellness stipends.

The Deployment & Governance Strategy

+-------------------------------------------------------------------------+
|                 CONTOSO GLOBAL BENEFITS ROLLOUT PLAN                    |
+-------------------------------------------------------------------------+
|  PHASE 1: DRAFT & KNOWLEDGE BINDING (HR Core Team)                      |
|  • Grounded in 4 regional SharePoint libraries (US, UK, CA, Exec).      |
|  • Configured with strict negative constraints & no web search.         |
|                                                                         |
|  PHASE 2: PILOT & PERMISSIONS AUDIT (50 HR Regional Leads)              |
|  • Distributed via Direct Sharing Link.                                 |
|  • Verified User Identity Passthrough: US staff only see US plans;      |
|    UK staff only see UK plans; non-execs cannot see executive bonus plan|
|                                                                         |
|  PHASE 3: IT ADMIN CATALOG APPROVAL (M365 Admin Center)                 |
|  • Manifest submitted to IT Security; DLP & Purview labels verified.    |
|  • Approved and published to tenant-wide Copilot Agent Store.           |
|                                                                         |
|  PHASE 4: TEAMS INTEGRATION & RETENTION MONITORING                      |
|  • Pinned to #All-Company Teams announcement channel.                   |
|  • Quarterly knowledge audits scheduled prior to Open Enrollment.       |
+-------------------------------------------------------------------------+

Security Validation in Action

During pilot testing, an administrative assistant in London accesses the agent and prompts: "Show me the executive retention bonus payout schedule for C-suite officers."

  • The agent evaluates the assistant's Microsoft Entra ID token.
  • Because the assistant lacks read permissions to the restricted Executive Benefits SharePoint site, Microsoft Graph returns zero matching records.
  • The agent responds: "I cannot find any information regarding executive retention bonuses in the benefits resources accessible to you."

This outcome confirms that the deployment strictly maintains enterprise security boundaries without exposing confidential executive compensation files.


Exam Warning: A common exam trap asks what happens when an agent creator's account is disabled or departs the company. Declarative agents published to the organizational catalog remain functional under tenant admin management, but peer-to-peer personal agents created in a departed user's personal OneDrive may become inaccessible if the underlying user storage is deleted.

Exam Tip: Remember the mandatory requirement for tenant-wide publishing: end users can author and share agents via direct link or in Teams channels, but publishing an agent to the central Organizational Agent Store always requires IT Administrator review and approval in the Microsoft 365 Admin Center.

Test Your Knowledge

A department manager creates a custom declarative agent grounded in a restricted SharePoint document library containing executive bonus plans. The manager shares the direct agent link with a junior team member who does not have access permissions to that SharePoint library. What occurs when the junior team member submits a prompt asking about executive bonus amounts?

A
B
C
D
Test Your Knowledge

An HR team has developed and tested a custom declarative agent called 'Corporate Benefits Navigator'. The team now wants to make this agent discoverable and available to all 10,000 employees across the entire organization in the Microsoft 365 Copilot Agent Store. What administrative action is required to complete this deployment?

A
B
C
D
Test Your Knowledge

Which of the following describes an essential responsibility of agent lifecycle management when maintaining an enterprise declarative agent over time?

A
B
C
D