Copilot Studio didn’t break Power Platform governance—it exposed how fragile it already was.

In this session, we dive into what’s really happening inside organisations as AI floods the platform: agents creating agents, copilots shipping code no one fully reviewed, and business demand accelerating faster than governance can react. The result?

Explosive Dataverse growth, unclear licensing boundaries, runaway token consumption, and connector sprawl across MCP, A2A, and everything in between.
These aren’t hypotheticals. These are real client situations—where “best practice” didn’t survive first contact with AI.

We’ll challenge the myths: that governance frameworks scale, that security models are keeping up, that cost is predictable, and that existing teams can simply “adapt.” You’ll see where things are working, where they’re being stretched to breaking point, and where—right now—there are no good answers.

We’ll also get practical. How do you govern environments when solutions aren’t the unit of risk anymore? How do you manage agents you didn’t design, code you didn’t write, and behaviour you didn’t explicitly configure? And how do you do all of this without doubling your team?

From Purview, Defender, and Agent 365 to Terraform and infrastructure-led controls, we’ll explore what’s actually helping bring order to the chaos—and what’s just giving a false sense of control.
If your governance model still assumes humans are in the loop, this session will challenge that assumption.

Key takeaways
• The uncomfortable reality of AI-driven Power Platform adoption in enterprise clients
• Where governance myths are actively causing risk, not reducing it
• How Copilot Studio is breaking traditional models for cost, licensing, and control
• Why token, connector, and agent sprawl is harder to manage than expected
• The new security risks introduced by autonomous and semi-autonomous agents
• Why solution-level governance is no longer sufficient—and what replaces it
• Practical patterns for governing AI without increasing headcount
• What tools genuinely help vs. what just looks good in architecture diagrams
• The gaps you cannot solve today—and how to mitigate them without false confidence