Enterprise AI is entering a new phase: instead of asking how to build AI agents, companies are now asking how to govern and operate them at scale. A year ago, almost nobody searched for "agent control plane." Then, in November 2025, search interest jumped roughly 25-fold worldwide when Microsoft unveiled Agent 365 at its Ignite conference and called it the control plane for agents. Since then, Salesforce, GitHub, Google, and IBM have all rolled out similar products using strikingly similar language, keeping search interest at roughly four times its pre-announcement level.

The reason for this sudden focus is straightforward: scale and complexity. Enterprises are no longer managing one or two AI agents. They're managing dozens or hundreds of agents built on different frameworks, connected to different tools, and governed by different policies. Once that happens, the challenge shifts fundamentally from agent creation to agent operations.

What Exactly Is an Agent Control Plane?

The term "agent control plane" comes from telecommunications, where it refers to the layer that handles signaling and instructions in older phone networks, separate from the layer that carries actual data. In enterprise AI, the concept translates directly: the control plane determines which agents can act, what they are allowed to access, how they interact with other systems, and how their behavior is monitored.

Think of it as a traffic control system for AI. Without it, dozens of autonomous agents operating independently across an enterprise create governance chaos. With it, organizations can apply consistent security policies, track what agents are doing, and ensure accountability when things go wrong.

The control plane is distinct from other AI infrastructure layers. Model Context Protocol (MCP) standardizes how agents reach tools and data; agent-to-agent protocols handle how agents talk to each other. But neither of those layers has an opinion about what an agent should be allowed to do once it starts acting. That's where the control plane comes in.

Why Are Large Enterprises Pulling Ahead in AI Agent Adoption?

The emergence of the control plane as a critical infrastructure component is coinciding with a widening gap in AI adoption between large enterprises and smaller organizations. McKinsey's latest research, based on responses from 1,719 participants across 97 countries, found a sharp divide in AI agent scaling. Among organizations generating more than $1 billion in annual revenue, 40% of respondents said they are now scaling AI agents, up from 27% the previous year. At smaller organizations, the figure is just 22%, unchanged from the prior year.

A similar divide is appearing with software coding agents. Around 20% of organizations overall are scaling the technology, compared with 31% of large enterprises. This gap matters because it signals that enterprise AI is becoming a two-speed race. Large enterprises have the resources, infrastructure, and governance frameworks to deploy agents at scale. Smaller organizations are still figuring out how to manage even a handful of them.

The financial implications are becoming clearer. McKinsey classifies just 6% of respondents as AI high performers, meaning they attribute at least 5% of earnings before interest and taxes (EBIT) to AI and describe its impact as significant. That proportion has not increased from the previous year, suggesting that simply spending more on AI doesn't automatically translate to better results. What distinguishes high performers is how differently they approach AI: they deploy a wider range of AI technologies, use AI for growth and innovation rather than focusing primarily on efficiency, and are 3.3 times more likely than other organizations to say they intend to fundamentally transform their businesses with AI over the next three years.

The Governance Challenge: Technology Alone Isn't Enough

Buying a control plane product is not the same thing as having a governance strategy. The technology can enforce policies, but organizations still need to decide what those policies are, who owns them, and how accountability works when agents make decisions or take actions. Someone still has to decide when an agent should ask a human for permission and what happens when things go sideways.

This is where many enterprises are stumbling. Governance has to be ongoing and operationalized, not treated as a one-time implementation exercise. In May 2026, a group of security companies released the Agent Control Standard, a vendor-neutral specification for runtime control, attempting to standardize how governance works across different platforms. More standards make the control plane more useful, not less, because the more interoperable the ecosystem becomes, the more important it is to have a common layer that can apply security policies, governance controls, and operational visibility across all components.

How Organizations Can Build Effective AI Agent Governance

• Define Clear Policies First: Before deploying a control plane, organizations need to decide what policies agents should follow, who owns those policies, and how they will be enforced across the enterprise.
• Establish Accountability Frameworks: Create clear processes for what happens when agents make decisions or take actions, including when human intervention is required and how escalation works.
• Operationalize Governance Continuously: Treat governance as an ongoing process, not a one-time setup. Monitor agent behavior, audit decisions, and update policies as new use cases emerge and risks evolve.
• Invest in Interoperability: Choose control plane solutions that support industry standards like Model Context Protocol and the Agent Control Standard to ensure governance works consistently across different agent frameworks and tools.
• Build Cross-Functional Ownership: Governance requires input from security, compliance, operations, and business teams. Establish clear ownership and decision-making authority across these groups.

The ROI Problem: Why Productivity Gains Aren't Translating to Profits

Enterprise AI has an unusual problem. Workers report significant productivity improvements from AI tools, but companies are struggling to turn those gains into measurable improvements in company-wide financial performance. McKinsey found that 80% of workers said AI has improved their individual productivity, and 50% said it helps them make better decisions. Yet only 37% of respondents said AI has made a positive contribution to their organization's EBIT, essentially unchanged from 2025.

This gap is creating pressure on organizations to demonstrate returns. One in five respondents said AI-related operating costs, including token costs, have constrained AI use. Most organizations still expect to increase their AI investment over the coming year, which means the pressure to show measurable returns is likely to grow as deployments become larger and more expensive.

Some vendors are responding to this challenge by embedding AI measurement directly into their platforms. WalkMe, a digital adoption platform, recently launched AI Authoring and AI Insights, two agentic AI capabilities that let professionals build in-app guidance and query usage analytics through plain-language conversation. The release targets a well-documented enterprise priority: 43.3% of AI decision-makers cite difficulty measuring business value as a top adoption challenge. By making it easier to measure how AI tools are actually being used and what impact they're having, WalkMe is addressing one of the core barriers to enterprise AI ROI.

The broader implication is clear: the next phase of enterprise AI may be less about who adopts AI and more about what happens after they do. Almost everyone is getting into the race. A much smaller group is beginning to pull away by redesigning how work gets done around AI rather than simply adding AI tools to existing processes.

" }