CIO's Guide to Governing AI Agent Proliferation

Corporate networks are rapidly filling with AI agents, creating a significant governance blind spot for leaders overseeing complex multi-cloud infrastructures.

As individual business units rush to adopt generative AI, CIOs are finding their digital ecosystems increasingly populated by fragmented and unmonitored assets. This trend mirrors the shadow IT challenges of the cloud's early days, but now involves autonomous actors that can execute business logic and access sensitive corporate data.

IDC forecasts that the number of actively deployed AI agents will surpass one billion by 2029—a staggering forty-fold increase from today's figures. In just the first half of 2025, agent creation skyrocketed by 119 percent. For enterprise leaders, the pressing challenge is no longer just building these agents, but locating, auditing, and governing them effectively across disparate platforms.

Salesforce is addressing this fragmentation head-on by enhancing its MuleSoft Agent Fabric with new automated discovery tools, designed to centralize the management of AI agents regardless of where they originate.

Automating Discovery

For security and operations teams, a lack of visibility remains the fundamental challenge. When marketing teams deploy AI agents on one platform while logistics teams build on another, central IT loses a consolidated view of the organization's digital workforce, making effective governance nearly impossible.

The updated MuleSoft architecture tackles this problem with 'Agent Scanners'. These tools continuously monitor major ecosystems—including Salesforce Agentforce, Amazon Bedrock, and Google Vertex AI—to automatically identify active agents. This eliminates the need to rely on developers for manual registration and reporting.

Finding an agent is just the beginning; compliance officers need to understand its purpose and scope. The scanners extract detailed metadata on an agent's capabilities, the large language models powering it, and the specific data endpoints it is authorized to access. This data is then standardized into uniform Agent-to-Agent (A2A) specifications, creating consistent profiles for assets across different vendor platforms.

Andrew Comstock, SVP and GM of MuleSoft, stated: "The most successful organizations of the next decade will be those that effectively leverage the full diversity of the multi-cloud AI landscape. The enhanced MuleSoft Agent Fabric provides the freedom to innovate on any platform while ensuring the unified visibility and control required to scale securely."

Governance and Cost Control for AI Agents

Unmanaged agents pose significant financial and security risks. For example, a banking CISO needing to verify a new loan-processing agent would traditionally have to manually track down documentation from development teams. Automated cataloguing allows security teams to instantly see which financial databases an agent can access and verify its authorization levels, ensuring they work with real-time data instead of outdated reports.

From a financial standpoint, visibility enables cost consolidation. Large enterprises often suffer from redundancy, with different regional teams independently purchasing or building similar tools. A multinational manufacturer, for instance, might have three separate teams paying for different summarization agents on three different platforms.

By using the MuleSoft Agent Visualizer to filter assets by function, operations leaders can quickly spot these overlaps. Consolidating them into a single, high-performing asset reduces redundant licensing fees and frees up budget for new, innovative projects.

Transitioning Successfully to an 'Agentic Enterprise'

Innovation frequently happens at the edges of an organization, where data scientists create custom tools outside of formal procurement channels.

The expanded Agent Fabric accommodates this by allowing the registration of "homegrown" agents and Model Context Protocol (MCP) servers via a simple URL. This is especially valuable in sectors like logistics, where teams might build internal tools for optimizing proprietary databases. Instead of remaining undocumented, these assets can be registered and made discoverable for reuse across the entire company.

Jonathan Harvey, Head of AI Operations at Capita, commented: "Agent Scanners will allow us to focus on innovation rather than inventory management. The assurance that every agent is automatically discovered and catalogued enables our teams to collaborate effectively, reuse existing work, and build smarter multi-agent solutions."

Similarly, AT&T is leveraging this framework to orchestrate agents across various customer interaction channels, including support, chat, and voice.

Brad Ringer, Enterprise & Integration Architect at AT&T, explained: "With the pace of AI development, MuleSoft Agent Fabric provides the essential framework we need to scale. It unifies and helps us orchestrate all the agents and MCP servers we're deploying in customer support, chat, and voice interactions. It's more than just a tool; it's a critical enabler for our entire forward strategy."

The shift to an "Agentic Enterprise" demands a fundamental change in how IT assets are governed. Relying on stale spreadsheets to manage integrations is no longer compatible with the speed of AI agent deployment.

Leaders must start with the assumption that their current inventory of AI agents is incomplete. Deploying automated scanning tools is essential to establish an accurate baseline. Once this foundation is in place, governance policies should require all agents—whether commercially purchased or internally built—to expose their capabilities and data access privileges in a standardized format like A2A to enable consistent monitoring.

Finally, executives can use the visibility these tools provide to audit spending, identify functional duplicates across cloud environments, and consolidate them to better control the Total Cost of Ownership (TCO).

As organizations transition from pilot programs to mass deployment, the key differentiator will not be the intelligence of any single agent, but the overall coherence and strategic management of the network connecting them all.

See also: Balancing AI cost efficiency with data sovereignty

Interested in learning more about AI and big data from industry leaders? Be sure to check out the AI & Big Data Expo happening in Amsterdam, California, and London. This comprehensive event is part of TechEx and is co-located with other premier technology events, including the Cyber Security & Cloud Expo. Click here for more details and registration information.

AI News is powered by TechForge Media. Discover more upcoming enterprise technology events and webinars here.