Companies Tripled Their AI Agents in a Year. Most Still Can't Prove It's Working.

Companies aren't just talking about AI agents anymore. They're deploying them by the dozen.

Salesforce's 2026 Agentic Enterprise Index found the average number of AI agents in production per organization grew from five in February 2025 to 13 by April 2026, according to ZDNET's review of the report. That's aggregated usage data from businesses running Salesforce's Agentforce platform, plus a survey of nearly 5,000 respondents across nine markets.

Companies now stand up a new agent in 1.9 days, down from four days in early 2025, a 53% drop, per ZDNET's reporting on the index.

These aren't dumb chatbots anymore either. BusinessWorld Online reported the average agent can now act on six skills, up from two at the start of 2025. During peak shopping periods, retail agents hit nine skills, a 350% jump, as businesses pushed agents to handle multi-step customer requests instead of one canned response.

Salesforce built its own metric to try to prove this is producing actual value: the Agentic Work Unit, or AWU, defined as one discrete task an agent completes. AWU output grew at a 15% compound monthly rate as of April 2026, according to Channel Dive. Retail agents alone accounted for 22% of total output, with their AWUs growing 18-fold over the analysis period.

Simons is one of the companies cited. The manufacturer sells across seven business units, employs 18,000 sellers, and gets about 2,800 unqualified inbound leads every week, BusinessWorld reported. Sales reps used to burn hours chasing leads with no idea of budget or timeline. Agentforce now handles that triage.

The catch: nobody agrees this is paying off yet

Salesforce is the company selling the agents, and Salesforce's own metric is the one showing the rosy growth numbers. That claim deserves scrutiny before accepting the ROI assertion at face value.

Keith Kirkpatrick, an analyst at Futurum Group, called the findings a "paradigm shift in AI deployment," Channel Dive reported. But the same article noted a competing survey from software company Aptean found less than half of organizations say AI is essential to their core work. And critics of the AWU metric itself argue it can measure activity without measuring actual revenue gained or costs cut, according to Channel Dive.

Salesforce's own Rob Schwartz insisted the numbers reflect real results: "We are seeing a measurable impact. There's a lot of noise out there, but our data is showing a positive ROI," he told Channel Dive. That's a vendor defending its own platform, not an independent auditor.

Outside the Salesforce ecosystem, the picture gets rougher. A Sinch survey found three-quarters of enterprises rolled back a customer-facing AI agent after deploying it, according to Customer Experience Dive. The top reasons: governance failures, customer data exposure, and hallucination or brand risk.

Julie Geller of Info-Tech Research Group didn't pull punches on why so many projects fail. "Taking shortcuts with AI cannot repair a fragmented customer journey, inconsistent knowledge, poor handoffs or a lack of ownership," she told Customer Experience Dive. "In fact, it can make that dysfunction faster, less visible and more difficult to unwind."

Antoine Nasr of Forethought AI Agents by Zendesk pointed to a leadership problem, not a technology problem. Executives see AI mandates as easy wins because "it's quite easy to get tools that generate tokens," he told Customer Experience Dive. "It's a lot harder to get tools that generate business value with those tokens." Customer Experience Dive reported that while 70% of CX practitioners say their organization has adopted AI, only a fraction are actually seeing return on investment.

Deloitte's research, covered by CIO Dive, backs up the skeptics further out. Tech leaders surveyed said their organizations are three to four years away from having even half their business processes redesigned around AI agents working autonomously with each other. Most companies today are bolting agents onto existing workflows instead of rebuilding those workflows from scratch, which is faster but doesn't deliver the technology's real upside, according to Deloitte's findings as reported by CIO Dive.

Widener, a source cited in the CIO Dive piece, framed the core difficulty plainly: agentic systems don't follow the same steps every time the way older automated tools do. "We're entering a different time where the technology and the agentic solution has the ability to think and take those outcomes and arrive at the best path of travel to get there without being told all the steps," Widener said. That unpredictability means someone still has to validate what the agent actually did, which is its own labor cost companies aren't always counting.

There's also a fair question about how far this technology can go before it hits a wall outside the enterprise back office. A new report on agentic commerce, cited by GlobeNewswire, found that while AI is reshaping how consumers discover and compare products, actual payment execution by autonomous agents remains limited. Existing payment authorization systems and identity verification aren't built for AI agents to complete transactions on their own, and consumer trust concerns around fraud and accountability are holding that back further.

None of this means the AI agent rollout is fake or that the adoption numbers are wrong. Salesforce's usage data is real, and the deployment speed is genuinely dramatic. But "we deployed 13 agents" and "we made money because of it" are two different claims. Right now most of the sourcing outside Salesforce's own report says companies have proven the first one, not the second.

The next data point to watch: whether Salesforce's fiscal year 2027 disclosures, expected in coming quarters, break out AWU growth against actual customer revenue or retention numbers, rather than presenting output volume as a stand-in for financial return.