A new report from Salesforce finds enterprises are rapidly expanding their use of AI agents but risk undermining those efforts unless they modernize how they connect systems, data and applications across the organization.

Salesforce’s 11th annual Connectivity Benchmark Report shows enterprises currently use an average of 12 AI agents. It projects that figure to grow 67% within two years as companies move toward what Salesforce describes as an “agentic enterprise,” where humans and AI agents work together across workflows. Yet half of those agents today operate in isolation rather than as part of coordinated, multi-agent systems, creating fragmented automation, governance risks and what IT leaders describe as the rise of “shadow AI.”

The findings are based on a survey of 1,050 IT leaders conducted between October and November 2025 by research firm Vanson Bourne with input from Deloitte.

In North America, 76 of the Top 2000 online retailers use Salesforce as their ecommerce platform, according to Digital Commerce 360 data. The Top 2000 is Digital Commerce 360’s database ranking North America’s largest online retailers by their annual ecommerce sales.

Salesforce findings on AI agent benchmarks

Nearly all respondents — 96%, said the success of AI agents depends on seamless data integration across systems. At the same time, only 27% of the average 957 enterprise applications are currently integrated. That underscored what respondents called a widening orchestration and governance gap.

“Agents are no longer experimental,” said Andrew Comstock, senior vice president and general manager of MuleSoft at Salesforce, in the report. “The real question for IT is how they are discovered, governed and orchestrated to work together as a system rather than as disconnected tools.”

The report shows companies are developing AI agents through multiple channels:

• 36% prebuilt SaaS agents.
• 34% embedded within enterprise platforms.
• 30% custom-built their agents in house.

That diversity is creating new management challenges for IT teams.

Concerns about introducing AI agents to workflows

86% of IT leaders said they are concerned agents could introduce more complexity than value without stronger integration frameworks. 42% cited risk management, compliance and legal concerns as top barriers to agentic transformation. A lack of internal AI expertise (41%), legacy infrastructure (37%) and difficulty integrating siloed data (35%) followed.

Data governance is emerging as a central issue. On average, 27% of enterprise APIs are considered ungoverned. And only 54% of organizations report having a centralized governance framework with formal oversight of AI and agent capabilities. Half of respondents (49%) identified cross-application data governance as a top integration challenge.

To close those gaps, IT leaders are turning to API-driven architecture as the foundation for multi-agent environments. 94% said future AI agent success will require IT architecture to become more API-centric, where APIs function as the connective layer between applications, data and AI systems.

One-third of teams said they are already using APIs to accelerate system integration. And 50% reported using APIs today to connect and govern AI capabilities.

Setting standards for agentic AI

Interest is also rising in emerging agent communication standards. Respondents said they are evaluating or planning to support protocols to enable agents to share context and collaborate securely. Those include:

• Agent Network Protocol
• Agent Communication Protocol
• Agent-to-Agent Protocol
• Model Context Protocol
• Universal Tool Calling Protocol

“This is an inflection point,” said Kurt Anderson, managing director and API transformation leader at Deloitte Consulting LLP, in the report. “Enterprises must move from simply deploying agents to operationalizing them at scale through sustainable and secure integration strategies.”

The report cites early adopters using integrated agent frameworks to move beyond experimental use.

AstraZeneca is using Salesforce’s Agentforce Life Sciences platform alongside MuleSoft integration capabilities to coordinate AI agents across field engagement, commercial operations and regional brands to improve how it interacts with health care professionals.

Enterprise intelligence firm r.Potential is combining Salesforce’s platform with multiple specialized agents and model context tools to generate executive-level workforce insights supported by a governed API foundation.

Despite rapid adoption, 96% of organizations reported barriers to using data for AI use cases. 40% cited outdated IT architecture caused by data silos and disconnected systems as the primary blocker.

The number of applications in enterprises also grew year over year from 897 to 957, compounding integration complexity.

The report concludes that without unified integration and governance, enterprises risk creating sprawling networks of intelligent tools that cannot effectively collaborate, limiting the productivity gains AI agents intend to deliver.

Sign up

Sign up for a complimentary subscription to Digital Commerce 360 B2B News. It covers technology and business trends in the growing B2B ecommerce industry. Contact Mark Brohan, senior vice president of B2B and Market Research, at mark@digitalcommerce360.com. Follow him on Twitter @markbrohan. Follow us on LinkedIn, X (formerly Twitter), Facebook and YouTube.