Vol. 31, The AI Control Plane Is Evolving
Forrester published an interesting piece this month identifying what they call three “functional planes” in enterprise agentic AI: one for building agents, one for orchestrating them inside business workflows, and a third—now emerging—for governing them across vendors and domains.
Industry is referring to that third layer the “agent control plane.” The framing is useful because it names something that’s been accumulating without a clear label.
Microsoft’s Agent 365 and AWS Bedrock AgentCore both shipped in recent weeks. Both use similar language. Both position themselves as governance and orchestration layers that sit outside the models and workflows they manage. The convergence suggests Forrester is describing something real, not just coining terminology.
What caught my attention is where this places the model in the overall picture. For a while, enterprise AI governance centered on models themselves—accuracy, training data, bias testing, prompt reviews, approval workflows. Those practices grew out of a period when models mostly answered questions and stayed in place. You could point to the intelligence, evaluate it, document it. That’s getting harder to maintain. In modern deployments, the model increasingly feels less like the decision-maker and more like a component inside a larger mechanism. What shapes behavior is everything surrounding it: how context assembles, which tools are available, how actions sequence, where authority gets granted.
Most of that logic never touches the model. It lives in orchestration.
The pattern should feel pretty familiar… databases taught us that access governance mattered more than query optimization, cloud platforms made identity and policy more consequential than server configuration. Control has a habit of drifting away from the most visible part of a system toward the layers that coordinate it.
If the Forrester framing holds, it raises practical questions about where governance attention should concentrate and whether current practices, built around static reviews and point-in-time approvals, will hold up when risk accumulates across steps rather than within them.
Worth watching.
References: • https://lnkd.in/ertXZZB3
hashtag #AlgorithmAndBlues hashtag #EnterpriseArchitecture hashtag #AIGovernance hashtag #AgenticAI