Salesforce Says Agent Deployments Doubled Year-Over-Year. Here's What the Number Doesn't Say
Salesforce's Agentic Enterprise Index reports that agent deployments more than doubled year over year, with activated agents up nearly 3x and average agent-creation time down 53%. Agentforce ARR hit $800 million, up 169% year on year, with 29,000 deals closed, up 50% quarter on quarter. The headline growth numbers are real and worth noting. The 83% autonomous resolution rate the same report cites is worth reading more carefully.
What "resolution rate" doesn't tell you
That figure is self-reported, drawn from Salesforce's own fiscal 2026 earnings materials, not independently audited, and the report doesn't fully define what counts as a resolution versus a partial handoff to a human. A vendor's own dashboard metric measuring its own product's success rate is a data point worth having, but it's not the same thing as an independent measurement, and it's the kind of number that's easy to define generously without technically being false.
Context from elsewhere in the same period cuts the other way. GitHub Copilot's own suggestion-acceptance rate sits at just 33%, a very different picture of "how often does the AI's output actually get used as-is." MIT NANDA research separately found that roughly 95% of generative AI pilots across industries show no measurable profit-and-loss impact at all. Both of those are also imperfect measurements, acceptance rate and P&L impact aren't directly comparable to autonomous resolution rate, but together they paint a more cautious picture than "deployments doubled" alone suggests.
What to actually ask a vendor citing growth numbers
Growth in deployments and ARR tells you the product is selling. It doesn't tell you the product is working at the rate the vendor's own success metric implies. Before treating an impressive-sounding resolution or success rate as proof a tool will work for your use case, ask how the metric is defined, whether it's independently verified, and what happens to the cases that don't count as a clean resolution. Deployment growth and actual measured impact are two different numbers, and vendors have every incentive to let you conflate them.