Photo provided by DepositPhotos
For the past two years, "AI agents" has been a phrase in search of a number. Vendors promised software that doesn't just answer questions but does the work. Skeptics asked whether anyone was actually deploying it.
New survey data puts a figure on it.
23% of companies are scaling agentic AI, McKinsey finds
McKinsey's latest State of AI survey, which gathered responses from 1,993 participants across 105 countries, found that 23% of organizations are already scaling an agentic AI system somewhere in the business. Another 39% have begun experimenting with agents. That puts 62% of companies at least testing software that can plan and execute multi-step work on its own.
People are also reading…
The distinction from earlier AI tools is concrete. A chatbot waits for a question. An agent looks at live data, decides what needs to happen next, and drafts the work.
The same survey found 88% of organizations now use AI in at least one business function, up from 78% a year earlier. AI use itself is no longer news. What the agent numbers show is growing evidence of the next phase: AI moving from producing text toward supporting operational decisions.
No function has crossed 10% yet
The same report puts a ceiling on the excitement. Most companies that are scaling agents are doing so in only one or two functions, and in no single business function do more than 10% of respondents report scaling them. Nearly two-thirds of organizations haven't begun scaling AI across the enterprise at all.
Agents have landed in specific corners of specific businesses, most commonly in IT and knowledge management, where use cases like service-desk automation matured fastest.
Manufacturing ranks among the top functions for AI cost savings
Follow the money and the same names keep coming up. Respondents most often reported cost benefits from AI in software engineering, manufacturing, and IT. Manufacturing's spot on that list makes sense once you look at what operational agents actually do all day.
Before any production run starts, someone has to reconcile orders against inventory and check material availability against vendor lead times. Someone has to catch the shortage that would stall a job three days in. It's slow, repetitive prep, and it's exactly what agentic systems were built for. Software like Fishbowl's AI operations specialist may already handle this on factory floors, working to check what's build-ready around the clock and drafting purchase and work orders from real demand data before shortages turn into missed ship dates.
Stripped of the buzzword, agentic AI looks like a purchase order that drafted itself overnight, waiting for a human to approve it in the morning.
High performers are 3 times more likely to be scaling agents
The sharpest split in the survey separates the companies getting results from the rest. Only 39% of respondents attribute any earnings impact to AI at all. But the small group McKinsey calls high performers, about 6% of respondents, are at least three times more likely than their peers to be scaling agents in most business functions.
The tools are widely available, and high-performing organizations appear more likely to deploy them in real business workflows.
30% of respondents expect AI to shrink headcount in the next year
The workforce data may be the strongest signal in the report. Across business functions, a median of 17% of respondents saw AI reduce headcount in the past year. Looking forward, a median of 30% expect a decrease in the next twelve months.
Companies are planning around agents, not just buying them. The 23% scaling figure will get quoted in pitch decks for the next year, but the gap between what AI did to staffing last year and what executives believe it will do next year says more about where business operations are headed. The survey suggests many respondents expect AI to have a greater effect on workforce size over the next year than it did during the previous year.
