MuleSoft Connect AI 2025: Live notes from the keynote
MuleSoft Connect AI 2025 was hosted in New York this month and the Computer Weekly Developer Network had a front row seat to listen to the keynote.
According to an opinion piece written by Andrew Comstock, SVP & GM, MuleSoft at Salesforce, disruptive technologies go hand-in-hand with transformations in business models.
For instance, he reminds us, cloud computing fundamentally moved IT cost structures from capital to operational expenditure.
“This shift empowered businesses to explore and innovate in new markets, products, and services. Agentic AI will go a step further, transforming the core processes and logic of how software operates and solves problems,” said Comstock.
But, to be truly effective, he says, AI agents can’t operate in a vacuum. They must be able to access and act upon data residing in various systems – from CRM and ERP platforms to legacy databases and modern cloud applications.
According to a MuleSoft-sponsored survey of over 1,000 enterprise IT leaders, the number of AI models used by organisations has doubled from last year, but an average of only 29% of enterprise applications are wholly connected inside organisations.
“As data silos multiply, agents are limited in their ability to retrieve information, automate processes and deliver meaningful outcomes,” said Comstock. “To build an agent-ready enterprise, robust integration and API capabilities are essential, with solutions like MuleSoft providing this critical foundation.
Once we connect systems, data and applications, MuleSoft promises AI agents will be able to retrieve information and take action, automate processes (and execute tasks across multiple systems, streamlining workflows) and extend reach as they interact with systems and data beyond their native environment.
Digital labour platform
Comstoc notes that MuleSoft for Agentforce extends the capabilities of Salesforce’s digital labour platform, Agentforce, allowing agents to securely perform actions across the entire business.
Eric Eyken-Sluyters, president of next-gen platform at MuleSoft hosted an opening session in the Connect AI keynote. He says that so many customers building agentic AI services today are questioning what kind of infrastructure they are basing those technology foundations upon. In preparation for what should be a multi-agent future, he thinks that organisations are now transforming business processes with these tools… but, crucially, they are looking to make sure they have visibility into those new services are they are built.
Crucial customisable composability
Comstock took over from his colleague to start talking about how agentic AI and the composability that is needed to make it happen is now becoming a reality. Reminding the audience that AI has the potential to change so much of business, Comstock says composability is not just about reuse, it’s thinking about how code can be refactored and used to make software teams more efficient and more functional with important support for emerging standards including MCP and A2A.
“We need to think about simple use cases for AI, we need to think about the end-to-end implications for a truly autonomous enterprise. We hear from customers every day who want to become and AI-agent-native enterprise… but (and this part is important) so many of the organisations at that point need rto question what they really means. Speed is the unprecedented advancement that all firms can have when it comes to agentic AI technologies now coming forward; being ‘first-to-market’ will be key for those that win with these new tools,” said Comstock.
MuleSoft insists it is now working to help orchestrate complex networks of agents to help them communicate and collaborate interactively.
Explaining how agent use cases actually work in production, MuleSoft senior product manager Mofeyi Oluwalana explained how admins can now use low-code tooling in MuleSoft for Flow, out-of-the-box connectors also allow business users to extract the data they are looking for when they need to build the integrations needed for workflows in modern business environments.
Importance of MCP Support
MuleSoft for Agentforce of course featured at this show. When we look at tools like Topic Center, API Catalog, Agentforce Connector… the company says it realises that there are so many agents out there (non Salesforce), which is why MuleSoft MCP support really came about in the first place. But agent-to-agent (A2A) communication is also important, so we must remember that agents must not just interact with one other, they should also be able to securely delegate tasks to other agents in the enterprise – and this is especially so if we look at agent-to-system interactions.
MuleSoft COO Ahyoung An spoke in deeper terms about agent-to-agent communication as they now start to work together to orchestrate complex processes. Where organisations might be worried about what data agents are accessing and what actions they are taking in the background, she says that MuleSoft has answers.
Flex Gareway is a flexible lightweight gateway to manage APIs.. and now that MCP and A2A Support for Flex exists, it can serve as an “agent control centre” and also work with API Manager to protect an IT stack with policies, which also works with the company’s API governance technology so that firms have an out of the box solution to gover all agent interactions.
Forward food for thought here spans subjects such as agent memory (agentic services can be empowered with both short-term and longer-term memory) and the question of how much memory we afford to agents based on their role and position in any given enterprise. Supercharging agentic software application development itself within comprehensively governed controls will be key, as will visibility into the agent estate that an organisation now builds.
Australian guitarist Steph Strings welcomed the MuleSoft Connect AI.
Andrew Comstock, SVP & GM, MuleSoft.