A global survey of 1,050 IT leaders finds 88% work for organizations that have, to one degree or another, already adopted artificial intelligence (AI) agents.
Conducted by the Mulesoft arm of Salesforce, the survey also finds nearly all (98%) plan to adopt AI agents, with organizations using, on average, 12 AI agents, which is expected to rise to 20 within two years.
A full 83% report that all or most teams and functions have adopted AI agents, with 19% of IT budgets expected to be allocated to AI agents and agentic transformation initiatives over the next 12 months.
However, a full 97% said they are encountering challenges, including risk management, compliance, security, and legal issues (42%), lack of internal expertise in AI/AI agent design (41%), legacy infrastructure or system incompatibility (37%), integration across siloed applications and data (35%) and business and IT misalignment (32%).
Additionally, 96% said they face difficulties leveraging data across the business for AI initiatives, with a quarter (25%) identifying data quality for AI or autonomous agents as the top data integration challenge. In general, 82% said data integration is one of the biggest challenges involving any type of AI adoption.
Half of respondents (50%) said AI agents operate in silos rather than as part of a cohesive multi-agent system. A full 96% agreed that the success of AI agents depends on seamless data integration across systems, with 86% noting that AI agents can introduce more complexity without proper integration.
At the moment, there does not appear to be a single preferred protocol for integrating AI agents. For example, the survey finds at 43% respondents report using both the agent network protocol (ANP) and the agent communication protocol (ACP), followed by the agent-to-agent (A2A) protocol at 40%, model context protocol (MCP) at 39% and the universal call tooling protocol (UCTP) at 34%, the survey finds.
That suggests the industry as a whole is still searching for the right set of standards to integrate AI agents, an issue that will make deploying AI agents increasingly more problematic, says Andrew Comstock, senior vice president and general manager for the MuleSoft arm of Salesforce. “Adoption without coordination is simply chaos,” he says.
Not surprisingly, nearly all respondents (99%) said their organization will use application programming interfaces (APIs) to automate and streamline processes, with 94% noting that AI agents will require a more API-driven IT architecture. Half (50%) are already using APIs to streamline and automate business processes for connecting and governing AI, but just over a quarter admitted that the APIs they have deployed are ungoverned. A total of 89% said there is room to improve the way APIs are managed within their organization.
Finally, the survey suggests that AI agents are being adopted quickly by application development teams within organizations. A full 94% said AI agents significantly improve development teams’ speed and efficiency, with 90% noting that developers prefer using AI-assisted coding tools over traditional methods for building integrations and APIs. Close to all (94%) said AI agents free developers up to focus on higher-value work.
While it’s still clearly early days so far as adoption of AI agents is concerned, the one clear thing is that the pace of adoption is now arguably rising faster than any other emerging technology in recent times.
