Essential Insights for Governing Copilot Agents | THR981
Tiffany Songvilay presents a practical framework for matching Copilot agent governance to the risks each agent introduces. The session connects data protection and operational controls with the decisions teams need to make before moving agents into production.
Governance should match agent risk
The session distinguishes between agents that answer questions, agents that support decisions, and autonomous agents that take actions. Governance requirements increase as agents gain access to sensitive information or the ability to write back to systems.
- Low-risk Q&A agents: Oversight starts with the information an agent can access and becomes more demanding when sensitive data is involved.
- Decision-making agents: Context awareness and data hygiene become important governance considerations.
- Autonomous, outcome-based agents: Write-back actions require a higher level of governance than answering questions alone.
Songvilay also covers Microsoft governance zones, spanning quick-assist scenarios through agents with strategic impact.
Data protection and context
Agent governance depends on the data and context available to the agent, not just its intended purpose. The session addresses challenges with unstructured data, SharePoint hierarchy, and centralized context, alongside sensitive data protection using Microsoft Purview, Copilot Studio, and data loss prevention (DLP) controls.
These considerations connect security and compliance requirements with the information agents use to answer questions, support decisions, or perform actions.
Intake, ownership, and production readiness
The session covers how citizen developers fit into an organization’s governance process, including:
- Intake processes for proposed agents.
- Risk classification and agent inventory practices.
- Clear ownership.
- The role of a Center of Excellence in assessing production readiness.
Together, these practices support oversight as agent adoption grows and help prevent agent sprawl.
Lifecycle oversight and scaling
Governance continues after deployment. Songvilay discusses lifecycle management and analytics that go beyond usage metrics to measure business outcomes.
The session also connects ownership, environment strategy, DLP controls, security, and compliance with production scaling. The aim is to enable agent development while maintaining appropriate control over sensitive data and agent actions.
Session chapters
- 00:00 — Introduction and governance by risk level.
- 01:18 — Low-risk Q&A agents and sensitive data.
- 02:40 — Decision-making agents, context awareness, and data hygiene.
- 04:24 — Unstructured data, SharePoint hierarchy, and centralized context.
- 05:23 — Autonomous agents and governance for write-back actions.
- 06:41 — Microsoft governance zones: quick assist through strategic impact.
- 08:30 — Citizen developers, intake, ownership, and agent inventories.
- 11:40 — Center of Excellence, risk classification, and production readiness.
- 14:00 — Lifecycle management and business outcome measurement.
- 16:56 — Ownership, environment strategy, DLP, compliance, and production scaling.