The CIO’s Governance Playbook

The proliferation of AI agents within corporate networks is creating a significant governance challenge for leaders overseeing multi-cloud infrastructures. As various business units rapidly adopt generative AI technologies, Chief Information Officers are increasingly finding their environments populated with fragmented and often unmonitored AI assets. This situation echoes the shadow IT issues experienced during the cloud adoption era, but with the added complexity of autonomous agents capable of executing business logic and accessing sensitive data.

Industry analysis projects a dramatic surge in AI agent deployment, with active agents expected to surpass one billion by 2029, representing a forty-fold increase from current levels. The first half of 2025 alone saw agent creation surge by an astounding 119 percent. Consequently, the primary challenge for enterprise leadership is shifting from building these agents to effectively locating, auditing, and governing them across diverse platforms.

Salesforce, through its MuleSoft Agent Fabric, is addressing this fragmentation by enhancing its capabilities with automated discovery tools. These innovations are designed to centralize the management of AI agents, irrespective of their origin or the platform on which they were developed.

### Automating Discovery for Unified Oversight

Visibility remains a critical hurdle for security and operations teams. When marketing departments deploy AI agents on one cloud platform and logistics teams build on another, effective governance becomes exceedingly difficult as central IT loses a consolidated view of the organization’s digital workforce.

MuleSoft’s updated architecture tackles this by introducing ‘Agent Scanners’. These tools continuously monitor major AI ecosystems, including Salesforce Agentforce, Amazon Bedrock, and Google Vertex AI, to identify deployed agents. Instead of relying on developers to manually register their AI creations, the system automates the detection process, providing a dynamic inventory.

Locating an agent is just the initial phase; compliance leaders require a deep understanding of its underlying logic. The scanners extract essential metadata, detailing the agent’s specific capabilities, the Large Language Models (LLMs) powering it, and the precise data endpoints it is authorized to access. This information is then standardized into Agent-to-Agent (A2A) specifications, creating a uniform profile for each AI asset regardless of the vendor or platform.

“The organizations that will thrive in the coming decade are those that can effectively leverage the full spectrum of the multi-cloud AI landscape,” stated Andrew Comstock, SVP and GM of MuleSoft. “The expanded capabilities of MuleSoft Agent Fabric empower innovation across any platform while ensuring the unified visibility and control necessary for scalable operations.”

### Governance and Cost Control for AI Agents

Unmanaged AI agents present both financial inefficiencies and significant security risks. Consider a Chief Information Security Officer in the banking sector. Under traditional operational models, verifying a new loan-processing agent would involve a laborious manual process of collecting documentation from development teams. Automated cataloging, however, allows security teams to instantly ascertain which financial databases an agent accesses and confirm its authorization levels without manual intervention, ensuring they are working with real-time, accurate data.

From a financial standpoint, enhanced visibility directly drives consolidation. Large enterprises often grapple with redundancy, where different regional teams independently procure or develop similar AI tools. A multinational manufacturer, for example, might find that three separate teams are each paying for distinct summarization agents on different platforms.

By employing tools like the MuleSoft Agent Visualizer to filter the AI estate by job function, operations leaders can pinpoint these overlaps. Consolidating these redundant assets into a single, high-performing solution can drastically reduce licensing costs and free up budget for new, strategic development initiatives.

### Transitioning to the ‘Agentic Enterprise’

Innovation frequently originates at the fringes, where data scientists develop bespoke tools outside formal IT procurement channels. The expanded Agent Fabric accommodates this by enabling the registration of “homegrown” agents and Model Context Protocol (MCP) servers via URL. This is particularly beneficial for sectors like logistics, where teams may build custom tools to optimize proprietary databases. Instead of remaining invisible, these internal assets can be registered and made discoverable for broader organizational reuse.

“Agent Scanners will allow us to concentrate on innovation rather than inventory management,” commented Jonathan Harvey, Head of AI Operations at Capita. “Knowing that every agent is automatically discovered and cataloged enables our teams to collaborate effectively, reuse existing work, and build more intelligent multi-agent solutions.”

Similarly, AT&T is leveraging this framework to orchestrate agents across a range of customer support functions, including chat and voice interactions.

“With the rapid pace of AI advancement, MuleSoft Agent Fabric provides the essential framework we need to scale our operations,” explained Brad Ringer, Enterprise & Integration Architect at AT&T. “It unifies and helps us orchestrate all of the agents and MCP servers we are developing for customer support, chat, and voice interactions. This is not merely a tool; it’s a significant enabler for our future endeavors.”

The transition to an “Agentic Enterprise” necessitates a fundamental shift in how IT assets are tracked, rendering traditional methods like managing integrations via outdated spreadsheets incompatible with the swift deployment cycles of AI agents.

Leaders must operate under the assumption that their current inventory of AI agents is incomplete and deploy automated scanning tools to establish a reliable baseline. Once this baseline is secured, governance policies should mandate that all agents, whether acquired or internally developed, expose their capabilities and data access privileges in a standardized format, such as A2A, to facilitate effective monitoring.

Finally, executives can utilize the visibility provided by these advanced tools to conduct thorough spend audits, identify duplicated functionalities across different cloud environments, and consolidate them to optimize the Total Cost of Ownership (TCO). As organizations move beyond pilot programs and into mass deployment, the key differentiator will not be the intelligence of individual agents, but the coherence and manageability of the network that connects them.