Agentic AI workforce is more than doubling year on year, says Salesforce
Salesforce reports a 160% year-on-year increase in AI agent deployments, with average agent numbers rising from five to 13 between February 2025 and April 2026.
Salesforce’s second annual Agentic Enterprise Index, covering data from February 2025 to April 2026, found that businesses more than doubled their agentic workforces, growing from an average of five agents to 13 over the period. The compound monthly growth rate (CMGR) reached 7%, with deployment times for new agents decreasing by 53% to an average of 1.9 days. The research tracked customers who maintained active agent deployments monthly, excluding those with intermittent usage.
Agents expanded beyond initial roles, increasing their average number of actions per account at a CMGR of 31%. Salesforce introduced its Agentic Work Unit (AWU) metric to measure agent productivity, reporting 734 million AWUs performed by April 2026, growing at 15% monthly. Analysts have questioned the metric’s relevance to business outcomes, though Salesforce emphasized its focus on task completion rather than token consumption.
The study highlighted agents operating across multiple cloud domains, underscoring the need for headless architecture to enable task execution and workflow triggering without traditional front-end interfaces. Within Salesforce, AI agent usage surged threefold in sessions between February 2025 and April 2026, with Slackbot saving employees an average of five hours weekly and adopted by 83% of staff.
Salesforce developed a Sophistication Index to assess agent complexity, ranking tasks from basic record lookups (Levels 1-3) to advanced functions like database updates (Levels 4-5). Manufacturing, financial services, and healthcare built more sophisticated agent networks than technology and retail. However, the most common and cost-effective starting point remains service use cases, with a shift toward action-oriented tasks replacing simple queries.