MuleSoft
APIs key to unlocking AI agent potential, says MuleSoft
Most Asia-Pacific IT leaders have either implemented or intend to deploy agentic AI in the next two years, but success will hinge on the implementation of APIs that AI agents rely on
A recent MuleSoft survey revealed that almost all (97%) IT leaders in the Asia-Pacific region have already implemented or plan to deploy artificial intelligence (AI) agents within the next two years, but harnessing the full potential of the technology will require a robust application programming interface (API) strategy.
That is according to Andrew Comstock, senior vice-president and general manager of MuleSoft at Salesforce, who noted that “it’s not just about having APIs, but about whether they truly embody your business processes”.
Speaking to Computer Weekly in a recent interview, Comstock argued that APIs must go beyond simple data access points and incorporate business logic to empower agents to perform complex tasks. “If the APIs are just access points to data, agents will just have data, but that doesn’t give them the context to complete the advanced tasks that might represent your business.”
He explained that most agents will require additional instructions that tell them what the APIs they are using do. In Salesforce’s agentic AI platform, Agentforce, for example, “action instructions” in the form of metadata provide natural language descriptions of what an API provides access to.
This need for context has led MuleSoft to introduce Topic Centre, which lets organisations annotate APIs with action instructions so they can be automatically ready for an agent to use. “That also gives you dual use, where APIs can be used by other systems in the traditional way, but they can also be used out of the box by applications like Agentforce to take advantage of their capabilities,” said Comstock.
While Agentforce can be used to build simple agents that answer questions about a document, for example, he said more advanced agents will be powered by APIs and business processes. For example, an AI agent that helps customers make hotel reservations will require it to look up the hotel’s room inventory, check if a customer has a potential reward status that can be applied to the booking, and process the transaction.
“Those are more complex business actions which are unique to the customer and need to function effectively,” said Comstock. “Those custom actions are increasingly being built with MuleSoft and utilised with applications like Agentforce.”
Read more about IT in APAC
- Global biopharmaceutical giant Takeda opens its first Asian Innovation Capability Centre in Bengaluru to speed up drug discovery and development.
- The Australian government is investing A$3bn to upgrade Australia’s fibre-to-the-node network, bringing faster internet speeds to hundreds of thousands of homes and businesses.
- StarHub is harnessing Red Hat OpenShift to power its hybrid multicloud platform, balancing control over critical infrastructure with the scalability and flexibility of public cloud.
- NCS is counting on its government experience, global expert network and cyber security capabilities to penetrate the enterprise IT market.
When it comes to managing and governing APIs alongside AI agents in Agentforce (for which Salesforce has built a lifecycle management tool for testing AI agents), Comstock said “you can protect an agent in the same way that you protect an API – it’s just changing what’s behind the scenes and the technology powering it”.
“That trend will continue as you want your AI agents to be working with the right data in the right context and for the right people to be accessing them, and that’s a great way to standardise that governance and protection,” he added.
Comstock acknowledged that the integration of APIs and AI agents is still an emerging area, but noted that companies like talent management firm Adecco are already leveraging MuleSoft and the Salesforce Data Cloud to centralise access across 40 systems, enabling recruiters to use Agentforce to speed up job placements.
“The pace at which AI agents are being adopted is dwarfing what we’ve seen for any other major technology,” said Comstock. “And so, the reality is that people are going to have to use the tools they have in place, like our API management and API governance capabilities, to protect their agents.”
