Salesforce says enterprise adoption of agentic AI is accelerating, with agents deployed faster and doing more work

Salesforce says enterprise deployment of “agentic” AI, software agents that can take actions rather than only answer questions, is expanding quickly. In its second annual Agentic Enterprise Index, the company reports that customers with agents running in production grew their agentic workforces from an average of five agents in February 2025 to 13 by April 2026, a 7% compound monthly growth rate.

The index tracks changes over five quarters, comparing customer activity between February 2025 and April 2026. Salesforce says it built the dataset around customers that had activated agents in production every month of the analysis period, and supplemented the findings with additional research from May 2026.

Speed is also improving. Salesforce says that by April 2026 it took an average of 1.9 days to deploy an agent into production, down 53% compared with the beginning of the report period. In practical terms, Salesforce is framing the trend as faster time from concept to operational rollout.

Once in place, Salesforce says the agents are increasingly taking on more work. Over the 15-month observation window, the average number of actions per account increased at a 31% compound monthly growth rate, suggesting agents are being used for a widening set of tasks after initial deployment.

Salesforce is attempting to quantify “work” performed by agents using a proprietary measure called an Agentic Work Unit (AWU), which is intended to capture actions completed rather than compute consumption such as tokens. The company says that as of April, Agentforce agents had performed 734 million AWUs, growing at about 15% each month. Analysts have criticized AWU as being disconnected from business outcomes, and Salesforce did not describe how AWU maps to customer revenue, cost savings, or service-level metrics.

During a media briefing, Salesforce executives said a key shift is that agents are expanding beyond their original scope. “These agents are expanding beyond their initial scope to really become cross-functional,” said Caila Schwartz, head of agentic commerce insights, describing the evolution from narrow assistants to systems that can coordinate across teams and processes.

The report also argues that agent architecture needs to be “headless,” meaning separated from traditional front-end user interfaces. Salesforce said its research showed agents increasingly operate across multiple cloud domains, and that decoupling the agent’s logic from user-facing screens enables it to process tasks, execute actions, and trigger workflows across environments.

Inside Salesforce, the company also pointed to its own usage as evidence of internal momentum. Joe Inzerillo, president of enterprise and AI technology, said sessions involving the Slack AI agent (Slackbot) rose threefold between February 2025 and April 2026. He added that Slackbot saves the average employee five hours per week and that 83% of employees had adopted it.

Still, the index leaves open questions about how broadly the results generalize. Salesforce’s measurements are drawn from customers with agents consistently in production each month, which may bias toward organizations that can sustain ongoing deployments and measurement. And while Salesforce presents sophisticated use cases and deployment trends, it does not provide a direct line in this report between agentic activity metrics such as AWUs and specific business outcomes for individual customers.

Looking ahead, the next announcement to watch is whether enterprises can translate faster deployment and higher action volumes into measurable improvements in service, sales operations, and back-office efficiency. Salesforce’s own framing points toward service as an early “best ROI” starting point for agentic systems, but companies will likely want clearer benchmarking on which agent categories, workflows, and governance approaches deliver the strongest real-world returns.