Enterprise AI adoption hits a practical wall
Half of U.S. workers now say they use AI, but uptake depends more on managerial support, workflow fit and perceived value than model access alone. Industry leaders also report a practical shift from chat interfaces toward agents that execute workflows, with companies calling out human‑system readiness as the bottleneck. (gallup.com) (x.com) (prnewswire.com)
Artificial intelligence is now common at work, but many companies are discovering that access to the tools does not make them routine. (gallup.com) Gallup said in January that 12% of employed U.S. adults use artificial intelligence daily on the job, about one-quarter use it at least a few times a week, and nearly half use it at least a few times a year. Its fall 2025 survey covered more than 22,000 workers. (wusf.org) The growth has been uneven. Gallup reported in December that 45% of workers used artificial intelligence at work at least a few times a year in the third quarter of 2025, up from 40% in the second quarter, while daily use rose only from 8% to 10%. (hrgrapevine.com) Gallup also said manager behavior is a dividing line. In organizations rolling out artificial intelligence, only 28% of employees strongly agreed in late 2025 that their manager actively supports its use. (forbes.com) That gap shows up in results. Gallup said employees who report manager support for artificial intelligence are 8.7 times more likely to strongly agree it has transformed productivity and 7.4 times more likely to say it gives them more opportunity to do their best work. (letsdatascience.com) Executives are also shifting the target. In a post highlighted by Forbes on March 31, Box chief executive Aaron Levie said companies now need people to “reimagine workflows for a world of agents,” not just hand workers a chatbot. (forbes.com) The distinction is practical. A chatbot answers questions in a window; an agent is meant to carry out a sequence of tasks across files and software, with rules for review and exceptions. (businesswire.com) Box has been building around that idea since September 11, 2025, when it introduced Box Automate as a system that breaks workflows into segments that can use artificial intelligence where needed and stay deterministic where risk is higher. (techcrunch.com) The work is hardest in the parts of business that run on documents, images and other material that does not sit neatly in database fields. Levie told TechCrunch that customer relationship management, enterprise resource planning and human resources systems were automated earlier because they handled structured data, while legal review and marketing asset work did not. (techcrunch.com) That helps explain why adoption can rise faster than returns. MIT Project NANDA said in a 2025 report that after reviewing more than 300 initiatives, interviewing 52 organizations and surveying 153 senior leaders, only 5% of integrated generative artificial intelligence pilots were extracting measurable value. (mlq.ai) The bottleneck is increasingly less about whether a model can draft, summarize or search, and more about whether a company has mapped the process, connected the data, assigned oversight and decided where humans still have to step in. (forbes.com) So the next phase of enterprise artificial intelligence looks less like a software rollout and more like an operations project. The workers already use the tools; the harder part is redesigning the job around them. (gallup.com)
