For years, one of the biggest challenges in enterprise software has been getting different applications to communicate with each other.
A company might use Salesforce for CRM, SAP for its operations, ServiceNow for IT support, and several other platforms for different business functions. Making all these systems exchange information reliably is where integration platforms like MuleSoft come into the picture.
But something interesting is happening.
We are moving towards a world where applications don’t just communicate with other applications. AI agents are also starting to interact with enterprise systems, retrieve information, trigger actions, and collaborate with other agents.
And this changes the integration conversation quite a bit.
MuleSoft seems to be taking this shift seriously, with several new capabilities introduced throughout 2026. Let’s look at what is changing and why it matters.
- From APIs to AI agents: What’s actually changing?
Traditional API integrations generally follow a predefined process.
An application sends a request, another application processes it, and a response comes back. Developers define the endpoints, authentication mechanisms, data structures, and business logic.
It works well because the process is predictable.
AI agents introduce a different approach.
Instead of simply following a fixed sequence of operations, an AI agent can interpret a request, decide which tools it needs, retrieve relevant information, and determine what action to take.
For example, imagine asking an AI assistant:
“Check the status of customer order 1050 and create a support ticket if it hasn’t been shipped.”
To complete this request, the agent might need to:
- Retrieve the order details from an ERP system.
- Check the shipment status.
- Evaluate whether the order is delayed.
- Create a support ticket if necessary.
The interesting part is that the agent needs access to multiple enterprise systems to accomplish this.
This is where existing API investments become extremely valuable. Instead of rebuilding every integration for AI, organizations can start making their existing services accessible to AI agents.
- MCP: A new way for AI agents to access APIs
One of the most interesting developments in this space is the Model Context Protocol (MCP).
MCP provides a standardized way for AI applications to discover and interact with external tools and services.
Think of it as a common communication layer between an AI application and the tools it needs.
Previously, connecting an AI application to different systems could involve building separate integrations and custom tool definitions.
MCP helps standardize that interaction.
MuleSoft has introduced MCP-related capabilities that make this particularly interesting for enterprise integration teams.
Enter MuleSoft MCP Bridge
Announced as generally available in MuleSoft’s 2026 mid-year release, MCP Bridge allows organizations to expose existing REST APIs as MCP servers.
This is significant because businesses don’t necessarily need to rebuild their existing integrations to make them accessible to AI agents.
For example:
Imagine an organization already has a MuleSoft REST API that retrieves customer information from an external CRM.
Using MCP Bridge, that existing API can be exposed through an MCP server. An AI agent that supports MCP can then discover and use the available functionality.
The existing API infrastructure can continue to handle authentication, policies, traffic management, and monitoring.
In simple terms, MuleSoft is helping bridge the gap between traditional API infrastructure and the growing AI ecosystem.
More details: https://blogs.mulesoft.com/news/mulesoft-26-mid-year-release-create-ai-experiences/
- Agent Broker: What happens when AI agents need to work together?
Connecting an AI agent to an API is one challenge.
But what happens when an enterprise has dozens of AI agents, each responsible for a different task?
One agent might handle customer support, another might manage inventory, and another might process financial information.
Making these agents work together reliably is a different problem altogether.
This is where MuleSoft Agent Broker comes in.
Agent Broker is designed to discover and route requests to suitable agents and tools.
Consider a customer support scenario.
A customer reports that an order has not arrived.
The workflow might involve:
- A support agent receives the customer’s request.
- It identifies that order information is required.
- An appropriate agent or tool retrieves the order details.
- Another agent checks shipment information.
- The support agent uses the results to determine the next action.
Instead of every agent needing direct knowledge of every other agent, Agent Broker helps coordinate these interactions.
Another interesting addition is Guided Determinism through Agent Script.
This addresses an important concern with AI: not every business operation should depend entirely on an AI model making decisions.
For example, an AI agent might be useful for understanding a customer’s complaint, but a high-priority security incident should follow predefined rules.
The combination of deterministic workflows and AI reasoning offers a more controlled approach.
- Agent Scanners: The problem of too many AI agents
As organizations experiment with AI, different teams often build agents using different platforms.
Some might use Salesforce Agentforce, others might use Amazon Bedrock, Google Vertex AI, or Microsoft Copilot Studio.
Over time, keeping track of all these agents becomes difficult.
Organizations need to know:
- Which AI agents exist?
- Where are they running?
- What systems can they access?
- Are they following organizational policies?
MuleSoft Agent Fabric addresses this through capabilities such as Agent Scanners.
These scanners automatically discover and register agents across supported platforms, giving organizations better visibility into their AI ecosystem.
This is an important development because introducing AI into an enterprise is not just about building more agents. It is also about managing and governing the agents that already exist.
- MuleSoft Vibes: Can integration development become more conversational?
Another interesting development is MuleSoft Vibes.
The idea is simple: instead of relying entirely on traditional development workflows, developers can use natural-language instructions to assist with building integrations.
This is particularly interesting for integration developers.
Think about a typical integration requirement:
“When a new customer is created in Salesforce, retrieve the corresponding customer information from an external system and update the Salesforce record.”
Traditionally, a developer would need to understand the requirements, configure the necessary connectors, build the flow, handle transformations, and test the integration.
AI-assisted development can help accelerate parts of this process.
However, this doesn’t mean developers suddenly become unnecessary.
Generated integration logic still needs to be reviewed. Data mappings must be correct, error handling must be considered, and security requirements cannot be ignored.
The developer’s role gradually moves towards designing, validating, and governing the integration rather than manually handling every repetitive development task.
- AI is also changing integration maintenance
Building an integration is only one part of the job.
Maintaining it is another.
Mule applications depend on Java and Mule Runtime versions. Keeping applications updated can involve checking compatibility, making changes, resolving issues, and running regression tests.
MuleSoft’s 2026 announcements also introduced Agentic Mule App Upgrades through MuleSoft Vibes.
The objective is to automate and accelerate Java and Mule Runtime upgrade activities.
This is an interesting direction because AI in software development is often discussed in terms of generating new code.
But maintaining existing applications is equally important.
In large enterprises, reducing the effort involved in maintaining hundreds of existing integrations could be just as valuable as building new ones faster.
- What does all this mean for MuleSoft developers?
I think this is where things get particularly interesting.
MuleSoft developers have traditionally focused on APIs, integrations, data transformations, connectors, error handling, and deployment.
Those fundamentals are not going away.
In fact, they are becoming even more important as AI agents start interacting with enterprise systems.
Developers may increasingly need to understand:
- How MCP servers expose existing API functionality.
- How AI agents discover and consume enterprise tools.
- How authentication and authorization work for agent interactions.
- How multiple agents communicate and coordinate.
- How to apply governance and monitoring to AI-driven workflows.
- How to validate AI-generated integration logic.
Learning MuleSoft today may therefore involve understanding more than just building flows and exposing APIs.
Understanding how those APIs become useful to AI agents could be an important additional skill.
Final thoughts
What I find interesting about MuleSoft’s direction is that it isn’t simply about adding AI features to an existing integration platform.
It is about adapting enterprise connectivity for a world where AI agents are becoming active participants in business processes.
APIs still matter. Integration still matters. Security and governance still matter.
But the way these capabilities are consumed is changing.
The next generation of integration developers may not just build connections between applications. They may also build the foundation that allows AI agents to interact with entire enterprise ecosystems.
And that makes the evolution of MuleSoft worth paying attention to.

