Salesforce Launches MuleSoft Agent Fabric: A Complete Breakdown

Salesforce is launching a new feature for MuleSoft which lets users more effectively control a wide range of AI agents – regardless of where they are built or operating.

The new solution, called MuleSoft Agent Fabric, is meant to turn fragmented, siloed agents into a “coordinated, trusted, and high-performing agent network”. 

The supposed problem Salesforce says is trying to solve is what it calls ‘agent sprawl’. This is when AI agents are built across teams, platforms, and vendors, meaning organizations are left with disconnected workflows, redundant automations, and compliance blind spots. In short: agents come from various sources and operate separately – and Agent Fabric unites them under one central authority. 

Let’s take a look at how exactly MuleSoft Agent Fabric works, the problem it aims to solve, and what this means for the Salesforce ecosystem. 

How Does MuleSoft Agent Fabric Work? 

Salesforce of course says that, with Agentforce, customers get a complete platform out of the box for building and deploying autonomous AI agents. 

MuleSoft Agent Fabric extends Agentforce to orchestrate with third-party agents that don’t interact through Agentforce, helping customers control their diverse ecosystem of agents with four main capabilities:

• MuleSoft Agent Registry: A central catalog where every AI agent or tool, including MCP and A2A servers, can be registered and made discoverable by developers or other agents, making every AI asset easy to find, reuse, and compose into workflows.
• MuleSoft Agent Broker: Powered by your LLM of choice and connected through A2A and MCP, is an intelligent routing service that organizes agents and tools into business-focused domains and dynamically routes tasks across them.
• MuleSoft Agent Governance: Provide enterprise-grade guardrails that apply security, compliance, and policy controls to every agent interaction, letting organizations scale AI adoption safely, helping ensure every action is consistent, secure, and aligned with enterprise and regulatory requirements.
• MuleSoft Agent Visualizer: This gives IT teams a dynamic map of their agent ecosystem, showing how agents connect, interact, and perform.

Architects can use MuleSoft Agent Visualizer to see a map of their agent network. Visualizer provides real-time insights into agent interactions, decision flows, and dependencies, helping teams optimize performance across multi-agent ecosystems.

MuleSoft Agent Broker, which teams can use to structure agents and MCP servers into business-focused domains. The Broker can navigate across these domains and route tasks to the best-fit agents and tools to quickly and accurately resolve incoming prompts.

MuleSoft Agent Registry will be generally available (GA) in October 2025. The beta of Agent Broker is available now, and will be GA in October 2025. Agent Governance (Flex Gateway for MCP & A2A) is available now. Agent Visualizer will be GA in October 2025.

The Bigger Picture and Use Cases

Salesforce hopes that, by providing a single place to orchestrate and govern agents, MuleSoft Agent Fabric will enable AI to solve complex, multi-step business problems across various industries. 

One example, provided by the cloud giant, is that of a mortgage application. Salesforce says that MuleSoft Agent Fabric can be used to “securely orchestrate” the various components of the mortgage application process.

A “Mortgage Assistant” agent, powered by Agentforce and embedded in a banking platform, can use the MuleSoft Agent Broker to “intelligently route a single customer inquiry across siloed SaaS and homegrown agents”, Salesforce says. 

Agentforce will capture the customer’s information and coordinate with an external Credit Check Agent for real-time credit checks, a DocuSign IAM Agent for secure document signatures, and a homegrown Compliance Agent to enforce regulatory policies. 

The MuleSoft Agent Registry makes each specialized agent easily discoverable for Agentforce, while MuleSoft Agent Governance helps to ensure every external agent interaction is secure and auditable, Salesforce says. 

The goal is to have a coordinated workflow that prevents critical handoffs from failing, reducing delays, errors, and significant compliance risks.

Salesforce Ben spoke to Andrew Comstock, Senior Vice President and General Manager of MuleSoft at Salesforce, about the new offering. 

We asked about potential security concerns when entrusting sensitive matters like a mortgage application to AI processes. Andrew stressed that agents and humans will continue to work collaboratively, emphasizing that full automation is not always the best approach for critical processes.

Andrew said: “We see agents and humans still working together, and I don’t think I’m going to take a mortgage application process and completely automate it so it’s fully agentic – I don’t think it’s the right answer to you… a mortgage application, that’s someone’s home, that’s someone’s life, that’s someone’s peace. You want the types of checks in place that I think you’re suggesting.

“The second part of it is… in many ways, mortgage applications are also black boxes today. You could imagine that someone who, as a customer, when you apply, you’re like, ‘I’m submitting all of this information, this entire packet of data. Wouldn’t it be nice if I could get some input on it… as opposed to it just instantly rejecting something?” 

Andrew also mentioned an anecdote, saying he recently spoke to a CIO of a “really large” tech company where there were worries about a data leak – which was discovered on the same day that an agent had been launched, and “everyone freaked out” thinking that the agent was responsible.

“The problem was they didn’t have enough controls on what data the agent was actually sharing to prove it wasn’t the agent,” Andrew said. “So they immediately had a team to start to figure out if the agent was broken. It turned out it wasn’t the agent at all. They put the guardrails in place that they needed. But because they couldn’t prove it, because they couldn’t even demonstrate to themselves, they spent a lot of cycles on it.”

Speaking about reliability, Andrew also mentioned an agentic use case where the AI can come back with immediate feedback for an issue, rather than having to go to a human.

He gave the example of a transportation company that has to collect driver’s licenses to register customers. This business saw their application rate “skyrocket” as soon as they had an agent say, “We can’t read the data on your license” due to too much glare, or whatever the issue was, speeding up processes.

Andrew said: “So if the answer is yeah, it missed some, it’s still they’re getting better results than they had without it. So I think that these are where these services can work together effectively to bring a better experience jointly and do things maybe that we weren’t doing previously.”

Salesforce also provides other examples of use cases for the new capabilities, including: 

• Real-Time Supply Chain Logistics: For example, an “Operations Agent” gets a prompt about a potential shipment delay, uses the MuleSoft Agent Broker to coordinate with a logistics partner’s “Fleet Agent” to re-route a delivery, while an “SAP Agent” updates inventory in real-time; then, a separate “Quality Control Agent” flags any potential issues at the warehouse.
• Partner and Employee Onboarding: Manage a complex onboarding process, for example, through the use of a “Partner Portal Agent” using the MuleSoft Agent Broker to coordinate with multiple backend systems, routing requests to a “Sales Agent” to assign a new partner ID, to a homegrown “IT Agent” for systems access, and to a “Knowledge Agent” to surface relevant articles and training.

Why It Matters (According to Salesforce)

As adoption of AI accelerates, with agents spreading across teams, platforms, and vendors, this brings in a new form of fragmentation – what Salesforce is calling “agent sprawl”.

The company says that, without the right foundation, unmanaged agents risk “creating chaos” instead of driving productivity, meaning it becomes harder to govern data, enforce security, and deliver consistent customer and employee experiences at scale.

Salesforce says that MuleSoft Agent Fabric can tackle this “agent sprawl” by transforming unmanaged AI agents into a secure and intelligent network. 

Salesforce said in a statement: “MuleSoft Agent Fabric provides a single place to register, orchestrate, govern, and observe every agent, regardless of where it was built. 

“Much like an air traffic controller ensures planes from different airlines take off, land, and share airspace safely, MuleSoft Agent Fabric acts as the backbone that connects and coordinates an enterprise’s disparate digital workforce, turning chaos into cohesion and transforming fragmented agents into a trusted, high-performing network.”

Joris Diependaele, Head of Integration and AI, Barco, said: “As AI agents become critical in processes that drive our day-to-day business operations, governance and transparency are non-negotiable. MuleSoft Agent Fabric provides the guardrails we need to adopt AI responsibly and securely, without slowing down innovation.”

Greg Shewmaker, CEO, r.Potential, said: “MuleSoft Agent Fabric gives us the foundation to orchestrate thousands of agents at scale, securely and in real time. It’s more than coordination – it’s the connective tissue for the next generation of enterprise intelligence. 

“By unifying disparate agents into one secure fabric, we help customers evolve from isolated automations to a fully orchestrated, adaptive ecosystem.”

Why It Might Not Matter That Much (At Least Right Now…) 

We asked Andrew whether ‘agent sprawl’ really is this big problem that needs solving right now. 

Figures from May this year show that, of Salesforce’s 150,000+ customers, 8,000 of those are currently leveraging Agentforce. 

Andrew said: “Is this a crisis for companies today? I don’t think it is, but what is interesting is that so many of the companies have gotten burnt – and I’m going to go back to cloud – there’s a sort of a pattern that almost every company that wasn’t built in the cloud natively felt which was, ‘Hey, we’re going to experiment with this, we’re going to give it to our developers, and we have this bill’, and it’s a couple hundred dollars, then a couple hundred thousand, all of a sudden it’s $10,000 a month and then it skyrockets from there.

“And we saw the same thing in APIs: People build a couple of APIs, it’s really great, makes teams effective, and all of a sudden it skyrockets from there. 

“And so I think in particular what I’m hearing from customers is they want to get in front of that, because if you think about governance on day zero, you’re designed [for scale] from the get-go.”

Andrew said Salesforce is taking an approach of bringing capabilities preemptively to make sure that CIOs and CTOs feel ready to deal with the scale of an agentic enterprise. This is a gamble for Salesforce. In their defense, sprawl is an emerging topic of interest around AI. But in technology it can be difficult to judge whether you’re getting ahead of a problem, or whether you’re falling victim to premature optimization. 

Clearly, given the pace of change in the current climate, Salesforce sees this as a risk worth taking. 

Final Thoughts 

While ‘agent sprawl’ might not currently be a plague upon the houses of Salesforce customers, Andrew makes a decent point about getting one’s house in order on ‘day zero’, so problems are prevented, rather than having to be addressed amid widespread panic later on when things go wrong.