Salesforce is moving deeper into agent-driven enterprise software, and MuleSoft is becoming central to that plan. The company’s goal is not just to help businesses build AI agents, but to manage how those agents reach data, call systems, and coordinate work across an organization.
In June, Salesforce released the third version of Agentforce, its toolkit for building agents inside the Salesforce ecosystem. The release adds a central control center and support for two agent protocols: Model Context Protocol (MCP) and Agent2Agent (A2A).
Why MuleSoft Matters To Salesforce’s Agent Strategy
MuleSoft is the technical backbone Salesforce is using for this shift. Salesforce acquired MuleSoft in 2018, and the platform has long focused on helping companies connect scattered systems through APIs. Those systems can include CRM tools as well as legacy on-premises infrastructure.
That role now has a new target: AI agents. If agents are expected to act across business software, they need reliable ways to reach databases, applications, knowledge bases, and workflow tools. MuleSoft’s existing integration position gives Salesforce a way to turn that connectivity into agent infrastructure.
Salesforce sees AI agents as a next step in office software. In that model, humans spend less time moving through screens manually, while more work happens automatically across different systems. The promise is not simply a smarter assistant, but a workflow that can detect a need, gather context, take action, and report an outcome.
MCP And A2A Split The Work
The two protocols in the new Agentforce release address different parts of the agent problem. MCP, developed by Anthropic, translates agent requests into system commands. In practical terms, that can mean actions such as querying a database or restarting a server.
A2A has a different role. The protocol started at Google and is now a Linux Foundation project. It lets agents assign tasks to each other, which matters when a workflow needs more than one specialized agent to complete a job.
The difference is straightforward:
- MCP helps an agent interact with systems.
- A2A helps agents work together.
That separation is important because enterprise workflows rarely live inside a single product. A customer agent from Anthropic, for example, could communicate with a product agent from OpenAI without developers having to create a custom protocol for that pairing.
But interoperability alone is not enough. An enterprise agent also needs boundaries. It has to know what it may access, which actions are permitted, and when a process should stop.
Flex Gateway Becomes The Control Point
MuleSoft’s Flex Gateway is designed to sit at the center of that control model. It decides which agents can access which data, what actions they are allowed to take, and when an interaction should end.
According to Salesforce, that setup lets companies turn existing APIs into agent-ready endpoints without rebuilding their infrastructure. This is a major part of the pitch: businesses can use systems they already have, while adding a layer that makes them available to agents under defined rules.
A typical workflow shows why the control layer matters. A monitoring agent might detect a critical error and alert a diagnostics agent. The diagnostics agent could check a knowledge base, then instruct a repair agent to restart the affected system. A fourth agent could then report the successful fix in Slack.
That sequence sounds simple, but it depends on several handoffs. Each agent needs the right access, the right instructions, and a clear stopping point. Without that, the same autonomy that makes agents useful can also make them difficult to manage.
Salesforce Wants The Interface Layer
Salesforce does not appear to be trying to build the single best agent itself. Instead, it wants to provide the "operating system" for agent-driven workflows. In this strategy, the key asset is the interface layer that agents use to reach business systems and each other.
The source article compares the approach to early cloud providers, which sold not only storage space but also the tools required to use it. The logic is similar here. If agents become a major way that work moves through software, the company that controls the interfaces can also influence the data flow.
That helps explain why MuleSoft is so important. Its original job was to connect systems through APIs. In the agent era Salesforce is describing, those same connections become the paths through which autonomous software can act.
The Market Is Interested, But Still Cautious
The technology remains early. Unlike classic software, agents interpret, learn, and react based on context. That makes them flexible, but it also makes their behavior harder to predict.
Several companies are already testing the idea. AstraZeneca is testing agent-driven processes for research and sales. Cisco Meraki is using MuleSoft’s "AI Chain" to connect its own language model with automated workflows for partner portals.
Even so, most efforts are still in beta. According to MuleSoft COO Ahyoung An, many customers are interested but hesitant because the technology is new and the risks are difficult to judge.
Early bug reports show why that hesitation exists. Some agents can get stuck in endless loops, while some processes never finish. These are not small concerns in enterprise IT, where automated actions may touch important systems and data.
MuleSoft is responding with training, entry-level pricing for smaller businesses, and a growing set of security policies. The two new protocols are scheduled for general release in July.
The bigger message is clear: Salesforce is betting that enterprise AI agents will need more than intelligence. They will need permissions, interfaces, monitoring, and limits. MuleSoft is the part of the Salesforce stack meant to make that possible.