Seth_Patton - Hopefully, this feedback is useful for both Microsoft and your customers globally. We are very keen to catch up via private chat to discuss further. The core blockers for us are dashboards and reporting, and reassignment of ownership (not changing Created by).
Copilot Agents (Agent Builder) – Risks, Blockers, and Limitations
We have identified a number of risks, blockers, and gaps with Copilot Agents built using Agent Builder during our initial pilot. These issues are grouped below into five main areas.
1. Administration & Reporting
From an administration and reporting perspective, we currently have almost no visibility or governance over agents created with Agent Builder.
Dashboards & Tools Assessed
- Viva Insights for Copilot
- Does not include Agent Builder agents.
- DSPM for AI (Apps & Agents in Purview)
- Does not include Agent Builder agents.
- Usage Reports – Microsoft 365 Admin Center
- Does not include Agent Builder agents.
- Agents & Connectors – Microsoft 365 Admin Center
- Provides only limited details (Name, Date Created, Publisher).
- No stats, adoption metrics, or differentiation between Agent Builder and Copilot Studio agents.
- This differentiation must be available in both the main table and the Excel export — currently a major gap.
- Export to Excel
- When no filtering is applied, it should export all agents.
- Currently, you must filter on Shared before exporting, otherwise Agent Builder agents are excluded.
- Date Created column
- Incorrectly shows the Last Modified date, not the true creation date.
- This is a bug that Microsoft must fix.
- Transfer ownership
- This is a MAJOR design flaw.
- Microsoft has confirmed that ownership transfer changes the immutable “Created by” property.
- This is unacceptable for auditing, legal, and compliance purposes.
- If I created an agent, the system must always reflect that — only ownership should change.
- This must be redesigned before rollout.
- Installed by / Sources display issue
- Always shows Installed by = 0 users (even if many have installed).
- Frequently fails to display the sources an agent is connected to.
- Message Center Post MC1134170
- The screenshot of the new dashboard still shows no differentiation between Copilot Studio and Agent Builder agents.
- The main table does not display Last Modified or Created by.
- More filters are required (e.g., filter by Owner, search beyond just Name).
Summary
- We currently have almost no admin, reporting, or governance options for Agent Builder agents.
- This is a major blocker for wider rollout.
- Unless Microsoft delivers significant improvements, we cannot scale Copilot Agents across the wider user base.
2. Ownership & Lifecycle Management
The ownership model for Agent Builder agents is a critical risk:
- Single-owner limitation
- Only the creator is the owner.
- No co-owners, no ability to reassign ownership.
- Consequences: agents become stranded when owners change role, go on leave, or leave the business.
- Microsoft’s proposed fix
- Ownership reassignment (GA planned for October).
- However, their current implementation overwrites the “Created by” field, which is unacceptable.
- A separate Owner attribute must be introduced.
3. Naming Confusion
Microsoft’s inconsistent naming conventions around Copilot agents are a significant issue:
- Sometimes full Copilot Agents are called Copilot Chat agents.
- Copilot Studio Agents are inconsistently defined — sometimes including both Copilot Studio and Agent Builder agents.
- Naming varies across:
- Message Center announcements
- Service announcements
- Roadmap items
- Documentation
- This inconsistency creates customer confusion and complicates governance and rollout planning.
- Microsoft must establish and use clear, consistent terminology across all channels.
4. Agent Limitations & Roadmap Gaps
There are several unresolved limitations and roadmap gaps that need urgent clarity:
- Co-authorship – No support for multiple authors/owners of an agent.
- Conversion to Copilot Studio – No in-place conversion option for Agent Builder → Copilot Studio agents.
- Use vs. Create split – No way to allow all users to use certain agents while limiting creation rights.
- Sharing controls – No ability to restrict how many people agents can be shared with or disable “Share with everyone.”
- Versioning – No version control (e.g., rollback to earlier versions).
- Supported data types – Microsoft indicated Agent Builder agents would soon support all file types, but this claim seems unlikely and requires confirmation.
- Knowledge source sharing
- Governance issue: if permissions are granted through an agent, they are not revoked when a user is later removed.
- This creates major compliance risks.
- Ownership/data access mismatch – Unclear what happens if ownership is transferred to a user who lacks access to the agent’s sources.
5. Technical Limitations & Bugs
Beyond governance, there are functional limitations and bugs:
- File output links
- Agents generate unusable sandbox links instead of valid file links.
- Microsoft acknowledged this but has provided no timeline for a fix.
- sandbox:/mnt/data/filename.extension
- False capability prompts
- Agents often state they can send emails or update systems, then retract saying they don't have that functionality.
- Frustrates users and reduces trust.
- Agents should at minimum support sending to Teams chat or send by email functionality.
- Endorsed vs. Test/Personal agents
- No way to distinguish between approved/endorsed agents vs. test/personal ones.
- Microsoft has said endorsement is “on the roadmap,” but with no details that we can find.
Overall Summary
- Administration & Reporting: Essentially non-existent for Agent Builder agents; major blocker to rollout.
- Ownership Model: Single-owner design creates significant operational risk; Microsoft’s fix is flawed.
- Naming Confusion: Inconsistent terminology across Microsoft channels confuses customers and hinders adoption.
- Agent Limitations: Co-authorship, conversion, sharing controls, and governance issues remain unresolved.
- Technical Gaps: Bugs (sandbox links, reporting inaccuracies) and lack of endorsement vs. test agent differentiation reduces trust and usability.