Enterprises now run an average of 12 AI agents, with the number projected to reach 20 by 2027, but half of those agents operate in complete isolation with no coordination between them. That’s the headline finding from Salesforce’s 11th annual Connectivity Benchmark Report, based on a survey of 1,050 IT leaders conducted by Vanson Bourne with input from Deloitte between October and November 2025. Salesforce published the findings in February; Belitsoft released a trend synthesis on April 6.
The isolation figure is the sharpest finding. Fifty percent of agents run as standalone tools rather than as part of coordinated, multi-agent systems, according to Salesforce’s own announcement. That fragmentation produces redundant automations, disconnected workflows, and what IT leaders in the survey described as “shadow AI” risks. Only 27% of the average 957 enterprise applications are currently integrated with one another.
Deployment Is Ahead of Governance
The overall adoption signal is clear: 83% of organizations now report that most or all teams have adopted AI agents, per Salesforce. The deployment split is roughly even across prebuilt SaaS agents (36%), embedded enterprise platform agents (34%), and custom-built in-house agents (30%).
But adoption is outpacing infrastructure. Andrew Comstock, SVP and GM of MuleSoft at Salesforce, framed the gap plainly: “Agents are no longer experimental. The real question for IT is how they are discovered, governed and orchestrated to work together as a system rather than as disconnected tools,” as reported by Digital Commerce 360.
The survey shows IT leaders are turning to API-driven architectures as the connective layer. Adoption or planned adoption of coordination protocols splits across Agent Network Protocol (43%), Agent Communication Protocol (43%), Agent-to-Agent Protocol (40%), Model Context Protocol (39%), and Universal Tool Calling Protocol (34%).
The Scale Projection
Futurum Group’s 1H 2026 Enterprise Software Decision Maker Survey, cited by Belitsoft, found that IT decision-makers naming autonomous agents as a top technology priority jumped from 13.0% to 17.1% in one year, a 31.5% increase. Planned agentic use cases cluster in cybersecurity (58.7%), sales and marketing (51.3%), and supply chain management (47.8%).
The cautionary number: just 11% of intended agentic use cases from the prior year made it to production, per Belitsoft’s synthesis. Concerns cited include risk tolerance, internal expertise gaps, legacy system incompatibility, and dispersed data storage.
By 2027, Salesforce projects the average enterprise will run 20 agents. Whether those agents are coordinated, or siloed, is the variable that determines whether doubling the fleet generates twice the value or twice the management overhead.