--- title: The Average Enterprise Runs 13 AI Agents Now, Up From 5. One Number in the Data Moved the Wrong Way. date: 2026-08-11T09:00:00Z author: Deepankar Bhadrasen url: https://truehorizon.ai/news/salesforce-agentic-index-skills-decline description: "Salesforce's Agentic Enterprise Index shows agent deployments surging. But headlines missed a key metric: average skills per agent have fallen since November 2025." --- # The Average Enterprise Runs 13 AI Agents Now, Up From 5. One Number in the Data Moved the Wrong Way. > Salesforce's Agentic Enterprise Index shows agent deployments surging. But headlines missed a key metric: average skills per agent have fallen since November 2025. Salesforce published the second edition of its [Agentic Enterprise Index](https://www.salesforce.com/news/stories/agentic-enterprise-index-insights-2026/) this week, and the headline numbers are genuinely striking: the average enterprise now runs 13 active AI agents, up from 5 in February 2025, and deploys a new one in 1.9 days instead of the better part of a week. Retail businesses running agents saw 4x higher online sales growth than the ones that didn't. Read one level deeper into the same dataset, though, and there's a number that moved in the wrong direction, and almost nobody covering the report led with it. ## What the index actually found The index draws on aggregated usage data from Salesforce's own Agentforce platform across thousands of businesses, plus a survey of 4,689 respondents fielded in May 2026. Agent counts grew at a 7% compound monthly rate over 15 months, nearly tripling. Deployment time dropped 53%. Weekly employee usage of agents tripled over the same period, and customer escalation rates stayed flat even as agent interaction volume scaled. On the numbers Salesforce chose to headline, agentic AI is compounding fast. ## The number that moved the wrong way The enterprise analyst outlet [diginomica](https://diginomica.com/agents-are-getting-more-skilful-and-more-numerous-according-salesforce-data-value-there-survive-ai) went back into Salesforce's own monthly breakdown and found that average skills per agent, the number of distinct tasks a single agent is actually capable of, peaked at 6.4 in November 2025. By the index's most recent data point, that number had fallen to 3.7. Agent count kept climbing through the same window. Put plainly: enterprises are deploying more agents, faster, and asking each one to do less. That is not the story a chart showing "13 agents, up from 5" tells on its own, and it changes what the growth actually means. ## A metric that's doing some quiet marketing work The index also introduces the Agentic Work Unit, or AWU: one discrete task completed by an agent. Agentforce agents produced 734 million AWUs in April 2026 alone, growing 15% month over month. That is a real usage signal. It is not a profit-and-loss outcome, and it is not audited by anyone outside the company that defined it, counts it, and sells the platform that produces it. Whether rising AWU volume is translating into actual business value is a separate question the index doesn't answer, and it's worth noticing that Salesforce's own data (more agents, each doing less) is exactly the pattern you'd expect if a usage metric were climbing faster than the value behind it. ## What this means if you're measuring AI ROI None of this means agentic AI isn't working. It means the count of agents you've deployed is one of the least useful numbers you can report to your board. Three things worth doing instead: - Track skill-per-agent and task-completion quality alongside deployment count. A larger fleet of shallow agents isn't the win a bigger headline number implies. - Ask whose interest a vendor's usage metric serves before you adopt it as your own KPI. A platform-defined activity count and a business outcome are not the same measurement, even when they're presented in the same chart. - Measure ROI in the currency your CFO actually uses: cost and revenue, not agentic work units, tokens processed, or any other activity metric a vendor invented to describe its own platform. ## Where TrueHorizon fits This is exactly the discipline we bring to every deployment. We've built agentic AI programs that get measured against real business outcomes, cost reduced, revenue moved, cycle time cut, not against a usage number a vendor's platform happens to be counting that quarter. That's not a checklist we're learning on your project. It's the expertise we bring to it. The Salesforce data is a genuinely useful signal that agentic AI is scaling fast. It's also a reminder that scale and value are two different questions, and answering the second one is where the real work is. If you're deploying AI agents and want to know whether the numbers you're reporting actually mean what they look like they mean, [take our AI readiness assessment](https://truehorizon.ai/assessment) before your next board update. Source: https://truehorizon.ai/news/salesforce-agentic-index-skills-decline