Survey: 72% enterprises run agentic AI

A social post on May 20 compressed several different enterprise AI findings into one claim: that 72% of enterprises already run agentic AI, while 60% still lack basic governance such as audit trails and overrides. The underlying numbers do point to a real pattern, but they do not appear to come from one public survey. The closest match comes from separate reports by Mayfield, TrueFoundry, MIT Sloan and Fast Company, each describing fast adoption and weaker controls from a different angle. ### Where does the “72%” figure actually come from? Mayfield’s January 2026 CXO Network survey is the clearest source for the 72% number. The firm said it surveyed 266 CIOs, CTOs, CAIOs, CISOs and CDOs, and found that 42% were already in production while 72% were deploying agentic AI in production and pilots combined. Zapier published a separate survey in December 2025 that also used 72%, but with a different definition. (mayfield.com) Zapier said more than 500 enterprise leaders were surveyed and that 72% of enterprises were “using or testing” AI agents; its breakdown said 40% had multiple agents in production and 32% were still in pilot or testing. That means the viral phrasing “72% in production” overstates what the public sources show. The available survey language supports 72% in production-plus-pilot, not 72% in production alone. (mayfield.com) ### What supports the claim that governance is lagging? Mayfield’s survey said 60% of respondents had an early-stage or no formal AI governance framework. The firm described a “speed vs. control” tension as enterprises moved agents into workflows faster than governance could follow. (zapier.com) TrueFoundry reported a similar gap on May 14, based on a survey of more than 200 enterprise AI leaders running agents in live production. (mayfield.com) It said 76% lacked unified logging across AI models and agent workflows, and 56% had no centralized control or governance layer. Those findings are not the same as “60% lack audit trails and overrides,” but they point in that direction. (mayfield.com) Unified logging, centralized controls and step-by-step tracing are core ingredients of auditability, and TrueFoundry said only half of enterprises had step-by-step tracing in place. That is an inference from the report’s control language, not a direct published percentage on “overrides.” (businesswire.com) ### Did MIT Sloan and Fast Company say leaders are getting this wrong? MIT Sloan’s Beth Stackpole wrote on May 18 that many organizations still fail to turn AI experimentation into business value. Nick van der Meulen of the MIT Center for Information Systems Research said companies “govern AI like legacy IT,” mistake productivity gains for enterprise value, and treat AI as a skill issue rather than an operating-model change. (businesswire.com) Fast Company’s April 7 article by Sudhir Prabhu said CIOs had moved past experimentation and were using agentic and generative AI operationally, while stressing governance, privacy and standards for scaling. A separate Fast Company piece published January 29 cited a Drexel University survey of more than 500 data professionals and said 41% of organizations were already using agentic AI in daily operations, while only 27% said their governance frameworks were mature enough to manage those systems effectively. (mitsloan.mit.edu) Together, those articles support the broader argument in the post: enterprise leaders are adopting agents faster than they are building controls around them. ### So what is the cleanest verified version of the story? The most defensible formulation is narrower than the social post. Publicly available sources show that 72% of enterprises were deploying or using/testing AI agents in production and pilots, depending on the survey, while separate surveys found governance gaps ranging from 56% lacking centralized control to 60% having early-stage or no formal framework. (fastcompany.com) (mitsloan.mit.edu) Cisco’s report with Omdia, based on 650 enterprise executives, adds the same direction of travel without using the same percentages. It said 87% believed agentic AI had reshaped strategic priorities and argued that governed, high-quality data, identity and security were prerequisites for scaling. Mayfield’s survey page, TrueFoundry’s May 14 report release, MIT Sloan’s May 18 article and Fast Company’s April 7 and January 29 pieces are the public documents that best anchor the claim set as of May 21, 2026. (mayfield.com) (cisco.com)