Salesforce tries to keep a lid on ‘agentic sprawl’ through MuleSoft Agent Fabric – News

Salesforce Inc. is attempting to tackle the growing challenge of artificial intelligence “agent sprawl” through the latest expansion of MuleSoft Agent Fabric, aiming to rein in the unchecked proliferation of autonomous systems operating across enterprises with limited oversight.

While some organizations have yet to deploy AI agents in any meaningful way, Salesforce says there are already hundreds of early adopters betting on the promise of agentic automation, and it believes that many more will do so in the coming years. According to data from International Data Corp., the number of deployed AI agents in the world will exceed 1 billion globally by 2029, representing a fourfold increase from 2025.

Agent sprawl is going to be a concern, Salesforce says, because many of them will be created by individual teams using different platforms, often without any consistent governance framework in place to ensure compliance and limit risk.

While this is still some way off in the future, Salesforce says there’s no time like the present to try and get on top of things, which is why it’s introducing a new “Agent Scanners” tool in its agentic orchestration platform, MuleSoft Agent Fabric. It’s designed to act as a single control plane for identifying, cataloging and managing AI agents across fragmented, multicloud information technology environments.

Agent Scanners is designed to automatically detect and register AI agents that run across diverse platforms including Salesforce Agentforce, Google Cloud Vertex AI and Amazon Bedrock. Once deployed, it will continuously monitor each connected environment and immediately flag when new or updated agents are set loose on it, capturing their technical context automatically.

Salesforce said this will help teams to replace the slow and error-prone process of tracking AI agents manually using spreadsheets and point audits. Should the organization’s finance team launch a new inventory forecasting agent built on Vertex AI, Agent Scanners will immediately pick up on it, without anyone needing to hunt for it through multiple cloud consoles.

The Agent Scanners don’t just identify AI agents, but also capture details via metadata on what actions each of the agents it finds is able to perform, the underlying large language models that power it, and what data access rights it has. MuleSoft Agent Fabric will then map this information to Google Cloud’s agent-to-agent protocol card specifications, so that organizations can create a consistent registry of all AI agents operating in their environments that explains how they reason and act. The MuleSoft Agent Registry will serve as a real-time source of truth for governance and security, and can also be used to register details of homegrown agents and Model Context Servers via simple URLs, to ensure internally-built tools aren’t excluded from the equation.

Salesforce said its ambition is to support the rise of the “Agentic Enterprise,” which is an operating model that sees AI agents work closely with humans across business functions such as inventory, customer service and logistics. With that in mind, the MuleSoft Agent Visualizer tool is also being updated. It can now show a consolidated view of an organization’s entire AI footprint at a glance, so teams can quickly filter through what agents they have deployed, based on their operating environments, capabilities or other metrics.

MuleSoft Senior Vice President and General Manager Andrew Comstock said he believes that the most successful organizations in future will be those that unlock agents across the full diversity of the multicloud landscape, but only if they have the right governance tools in place. “The expanded capabilities of MuleSoft Agent Fabric give you the freedom to innovate across any platform while maintaining the unified visibility and control needed to scale,” he promised.

Featured image: News/Meta AI