For the past year, the enterprise AI conversation was dominated by a simple, vanity-driven metric: how many chatbots can a company launch? It was a race for volume. However, the Salesforce Agentic Enterprise Index, which released its second edition on August 7, 2026, suggests that the industry has finally moved on. The focus is no longer on the sheer number of agents, but on the action-to-output ratio, which is currently growing at a 15% compound monthly growth rate (CMGR). This ratio is a far more useful way to track progress. It measures the efficiency of an agent’s execution—specifically, how many meaningful tasks are completed per unit of effort. By shifting the lens from deployment counts to actual output, we get a clearer picture of what these systems are doing for the bottom line.
The data from the 2026 report highlights a significant shift in how companies are using AI. By April 2026, organizations had performed 734 million Agentic Work Units (AWUs). Perhaps more telling is that the number of activated agents per organization has tripled, moving from 5 to 13. This suggests that businesses are moving past the experimental pilot phase and into functional, multi-agent workflows.
As Joe Inzerillo, Salesforce’s EVP and Chief Technology Officer, noted, the industry is moving from passive chatbots and predictive models to execution-driven agents. This transition is showing up in the numbers. Retailers using these systems saw 4x higher sales growth—8% compared to the 2% industry average—year-over-year. Internal adoption is also climbing, with employee weekly usage tripling, suggesting that staff are finding real value when agents solve problems rather than just deflecting queries.
Specific implementations illustrate this shift. Pandora, for instance, deployed an agent named Gemma to handle routine support. Gemma now manages 60% of those tasks, which contributed to a 10% increase in Net Promoter Score (NPS). Similarly, PenFed has integrated agents named Ace and Echo to handle specific banking tasks. These are not simple automated scripts; they are agents with expanded skill sets, which have grown from an average of two to six per agent.
Despite these gains, it is important to read these figures with a critical eye. The report includes a significant caveat: survivorship bias. The data only includes businesses that kept agents in production every single month from February 2025 to April 2026. This means the index reflects the performance of the most successful, persistent adopters, rather than the average experience of every company that attempted to launch an agent.
The reality of current technology is also tempered by the escalation rate, which remains steady at 32%. Even with more sophisticated agents, nearly a third of interactions still require human intervention. This serves as a necessary check for anyone expecting total, hands-off automation. As Shree Reddy, EVP of Product Management at Salesforce, explained, the goal is to pair robust governance with a unified platform to safely deploy these multi-action agents.
Financial data further underscores the momentum behind this shift. Salesforce reported that its Agentforce Annual Recurring Revenue (ARR) has hit $800 million, a 169% increase year-over-year. This growth indicates that enterprises are willing to invest heavily in tools that demonstrate clear, measurable efficiency gains.
The 2026 index suggests that the “agentic” label is finally being tested against performance. The focus on the action-to-output ratio forces both vendors and buyers to justify the technology based on what it actually achieves. If the current CMGR holds, we should expect to see even more specialized agents in the coming year.
The real test for the industry will be whether that 32% escalation rate can be lowered without sacrificing the quality of the output. For now, the data shows that the most successful enterprises are those that treat agents as a core part of their operational infrastructure, rather than just a shiny new feature.