Catalogue › Copilot Studio › studio-deployment
Studio deployment
Publishes a released Copilot Studio agent to its channels - Teams and Microsoft 365 Copilot, websites, custom apps and others - configuring per-channel authentication, sharing to the approved audience, admin approval and availability, channel-specific behaviour and a staged rollout with a rollback route. Use when an agent has passed its release gate, when it is being added to a new channel, or when users cannot find or open it.
- Department
- Copilot Studio
- Lifecycle
- Launch, Scale
- Maturity
- adopter, builder, scaler, ai-enterprise
- Risk tiers
- 1, 2, 3, 4
- Platforms
- copilot-studio, m365-copilot, m365-agents, power-platform
- Editions
- starter, scale, enterprise
- Reviewers
- security-privacy
- Status
- draft
- Copilot Cowork
- Copilot Studio plugin for Copilot Cowork, or this skill on its own: Download SKILL.md
- Licence
- CC BY 4.0, by Simon Foote
Purpose
An approved agent delivers nothing until the right people can find and use it in the place they work. This skill takes a Copilot Studio agent that has passed its release gate and publishes it to its channels: configuring each channel, matching sharing to the approved audience, getting any admin approval, and rolling out in stages with a way back. Deployment usually goes wrong when an agent is shared more widely than approved because it was easier, when a website channel is enabled without checking authentication, or when users cannot find an agent because nobody arranged its availability in Teams and Microsoft 365 Copilot.
When to use
- An agent has passed
m365-agents:agent-release-readinessand needs publishing to its channels. - An existing agent is being added to a new channel or a wider audience.
- Users report they cannot find, install or open the agent.
- A staged rollout or a rollback of a published agent is needed.
When not to use
- Deciding whether the agent is ready to release - use
m365-agents:agent-release-readiness. - Moving the agent between environments with solutions and pipelines - use
copilot-studio:studio-alm. - Setting the organisation's release policy by risk tier - use
platform-governance:ai-release-policy. - Planning launch communications and user onboarding - use
copilot-adoption:launch-plan.
Questions to ask first
- Which audience did the release gate approve, and which channels were tested?
- Will users meet the agent in Teams and Microsoft 365 Copilot, on a website, in a custom app or elsewhere?
- Does each channel support the agent's authentication and single sign-on as designed?
- Does making the agent available to users require approval from Microsoft 365 or Teams administrators, and who are they?
- Will rollout be staged, and what would trigger pausing or rolling back?
Required inputs
- The release decision and approved audience from
m365-agents:agent-release-readiness, with theagent-release-checklist. - The production version imported through
copilot-studio:studio-alm. - The authentication design and per-channel sign-in test results from
copilot-studio:studio-authentication. - Security groups for the approved audience and contacts for Microsoft 365 or Teams administration.
If the release decision is missing, do not publish. If a channel was not tested, publish to the tested channels only and record the rest as pending.
Decision framework
| Channel | Use when | Check before publishing |
|---|---|---|
| Teams and Microsoft 365 Copilot | Staff audiences in their daily tools | Admin approval and availability rules, single sign-on, audience security group |
| Organisation website or intranet page | Staff or customers on a web page | Authentication matches audience; embedding secured; no internal tools on anonymous pages |
| Custom or mobile app | The agent is part of an existing application | App owner, authentication handshake, support route |
| Other messaging or contact centre channels | The audience already uses them | Data leaving the tenant, hand-off behaviour, channel-specific formatting |
| Rollout stage | Audience | Move on when |
|---|---|---|
| Pilot | A named pilot group | Threshold met in real use, no zero-tolerance incidents |
| Expanded | The first business units or roles | Support load and analytics stable |
| Full | The approved audience | Release review confirms readiness |
Never share beyond the approved audience. Widening the audience is a new release decision.
Practitioner approach
Confirm the decision before touching the agent. Read the release record: the approved audience, the channels tested, the risk tier and any conditions. Deployment implements a decision; it does not make one.
Configure each channel deliberately. For Teams and Microsoft 365 Copilot, prepare the agent's name, description, icon and availability so users can recognise it, and work with Microsoft 365 or Teams administrators on approval and how the agent is made available to the audience. For websites and custom apps, confirm that the authentication configured for that channel matches the audience; an internal agent embedded on a public page is a data exposure, not a deployment detail. Check how each channel renders the agent's messages, cards and citations, and adjust where a channel does not support something the design relies on.
Share to security groups, not to individuals or "everyone". Sharing to the group that matches the approved audience keeps access aligned with the release decision and makes later changes auditable. Check sharing limits and admin controls in the current source records before planning large audiences.
Roll out in stages. Start with a pilot group drawn from the approved audience, watch analytics, support contacts and feedback for an agreed period, then widen. Agree pause and rollback triggers in advance: a zero-tolerance failure, a spike in failed actions, or a pattern of wrong answers. Rolling back may mean unpublishing a channel, narrowing sharing or re-importing the previous version through ALM; know which, and who can do it, before you start.
Coordinate with launch. Users need to hear about the agent where they work, with examples of what it does well and where to get help. Hand the timing and channel details to copilot-adoption:launch-plan so communications match availability.
After each stage, record what was published, to whom, in which channels and when, in the agent's release record and the solution inventory. Verify as an ordinary user in each channel that the agent can be found, signs in correctly and answers within its scope.
Outputs
- A channel configuration record per channel: settings, authentication, appearance, admin approvals and test result.
- A staged rollout plan with audiences, dates, pause and rollback triggers and responsible roles.
- Sharing records showing the security groups used and their match to the approved audience.
- Updates to the
agent-release-checklistand the solution inventory after each stage.
Microsoft platform considerations
Channels, admin approval routes, availability in Microsoft 365 Copilot and Teams, sharing limits and channel features change often. Check ms-copilot-studio-guidance and ms-agents-extensibility for current channels and behaviour, ms-copilot-studio-security-governance for admin and sharing controls, and ms-copilot-studio-zoned-governance for which zones may publish where.
Adoption considerations
Availability without communication produces no usage. Align deployment stages with launch messages, give the agent a clear name and description users will recognise, and make sure support knows the agent exists before users do.
Knowledge considerations
Not applicable to channel configuration itself, because knowledge is fixed at release; check after publishing that citations and links open correctly for users in each channel.
Security and privacy considerations
Sharing never exceeds the approved audience, anonymous channels never reach internal data, and data leaving the tenant through a channel is reviewed. Channel changes that widen exposure go back to security-privacy:data-exposure-review.
Responsible AI considerations
Each channel must carry the agent's AI disclosure and route to a person. Public-facing channels deserve particular care because users may not expect to meet an AI agent; confirm disclosure with copilot-studio:conversation-design.
Evaluation requirements
Verify in each channel as an ordinary user: discovery, sign-in, a sample of in-scope and out-of-scope questions and one action where the agent has tools. Monitor the pilot stage against the release threshold before widening.
Value measures
Track reach within the approved audience, time from release decision to availability, and pilot-stage outcomes, owned by the agent owner and reported at agent and audience level.
Scale notes
Starter
Publish to Teams and Microsoft 365 Copilot for a security group matching the approved audience, pilot with a small group first and keep the previous version for rollback.
Scale
Standard channel configurations, an agreed admin approval route, staged rollouts by default and release records linked to the solution inventory.
Enterprise
Formal deployment windows aligned with change management, staged rollouts across regions, automated checks that sharing matches approvals and portfolio-level reporting of channel coverage.
Hand-offs
- To
copilot-adoption:launch-planwith channel availability and timing. - To
copilot-studio:studio-analyticsto monitor the pilot and each stage. - To
copilot-studio:studio-almfor rollback by re-import. - To
copilot-studio:studio-agent-lifecycleonce the agent is fully live.
Reviewer gates
From risk tier 2, security-privacy confirms that sharing and channels match the approved audience and that anonymous or external channels cannot reach internal data. From tier 1, any public website channel needs the same check. Sharing beyond the approved audience is a blocking finding.
Failure modes
- The agent is shared with everyone because creating a security group felt slow.
- An internal agent is embedded on a public page with authentication disabled.
- The agent is published but not made available in Teams and Microsoft 365 Copilot, so nobody can find it.
- No rollback was planned and a faulty release stays live while the team improvises.
Sources
references/sources.md in this skill lists the source records behind this skill, what each is authoritative for, how volatile it is and what you may do with it. Product availability, licensing, limits and preview status change often, so check the current source before stating them, and cite what you used.
Never
- Publish an agent without a recorded release decision.
- Share an agent beyond its approved audience, or with everyone by default.
- Enable a public or anonymous channel for an agent that can reach internal knowledge or tools.
- Start a rollout without agreed pause and rollback triggers.
Source records for this skill
Microsoft Copilot Studio guidance documentation
- Source id:
ms-copilot-studio-guidance - Publisher: Microsoft (product-documentation)
- Link: https://learn.microsoft.com/en-us/microsoft-copilot-studio/guidance/
- Authoritative for: Planning, adoption maturity, governance, testing, ALM and measurement guidance for agents.
- Volatility: medium; review every 90d; last reviewed 2026-10-29
- What you may do: Read and cite. Summarise in your own words; do not reproduce the text.
Copilot Studio security and governance
- Source id:
ms-copilot-studio-security-governance - Publisher: Microsoft (product-documentation)
- Link: https://learn.microsoft.com/en-us/microsoft-copilot-studio/security-and-governance
- Authoritative for: Security and governance controls for Copilot Studio agents, including data policies, environment routing and maker controls.
- Volatility: high; review every 30d; last reviewed 2026-10-29
- What you may do: Read and cite. Summarise in your own words; do not reproduce the text.
Ways to extend Microsoft 365 Copilot
- Source id:
ms-agents-extensibility - Publisher: Microsoft (product-documentation)
- Link: https://learn.microsoft.com/en-us/microsoft-365/copilot/extensibility/overview
- Authoritative for: The extensibility options for Copilot, including agents, connectors, skills and tools, and how to choose between them.
- Volatility: high; review every 30d; last reviewed 2026-10-29
- What you may do: Read and cite. Summarise in your own words; do not reproduce the text.
Implement a zoned governance strategy (Copilot Studio guidance)
- Source id:
ms-copilot-studio-zoned-governance - Publisher: Microsoft (product-documentation)
- Link: https://learn.microsoft.com/en-us/microsoft-copilot-studio/guidance/sec-gov-phase2
- Authoritative for: Segmenting agent-building into governance zones by purpose and risk, and the controls available at each level.
- Volatility: medium; review every 90d; last reviewed 2026-10-29
- What you may do: Read and cite. Summarise in your own words; do not reproduce the text.
Credit
This skill was created by Simon Foote as part of CopilotForge, licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Credit the author when you share or adapt it.