AI has crossed from experiment to operating reality faster than most executives can rewrite their org charts. The new shock is not that companies are dabbling with chatbots. It is that AI adoption is already broad, uneven, and strangely underpowered: a major NBER working paper surveying nearly 6,000 senior executives across the United States, United Kingdom, Germany, and Australia finds that 69% of firms actively use AI, while most still report little measurable impact on jobs or productivity so far.
That gap should make leaders uneasy. The tool is in the building. The transformation is not. The next competitive divide will not separate companies that have AI from those that do not. It will separate firms that convert AI into better workflows, faster decisions, and sharper execution from firms that merely add another tab to the browser.
Adoption Is Running Ahead Of Impact
The headline numbers looks revolutionary. In the NBER study, 78% of U.S. firms, 71% of U.K. firms, 65% of German firms, and 59% of Australian firms reported using at least one AI technology, with text generation through large language models leading the pack. Yet the same research finds that more than nine in ten executives reported no employment impact over the past three years, and 89% reported no productivity impact.
That is the paradox of the AI survey: adoption is widespread, but organizational change is still shallow. Executives themselves average only about 1.5 hours of AI use per week, according to the same survey, which helps explain why a technology can be everywhere in principle and still marginal in practice.
Other evidence points in the same direction. The U.S. Census Bureau’s business survey data shows overall AI usage among U.S. businesses hovering between 17% and 20% from December 2025 to May 2026, with larger firms using AI more heavily.
The Federal Reserve’s review of Census data similarly found that about 18% of U.S. firms had adopted AI by the end of 2025, while noting rapid growth before a late-2025 methodology change in the survey.
That discrepancy is not necessarily a contradiction. Surveys define AI differently, ask different people, and sample different firms. A senior executive may classify machine learning, visual content creation, and LLM use as AI across the business, while a Census questionnaire may capture narrower implementation.
The common message is clear enough: larger, more productive, and higher-paying firms are moving first, while smaller and older organizations risk letting AI become another technology that deepens existing advantages.
Productivity Gains Need Management More Than Magic
The best evidence for AI’s potential remains powerful. A business productivity study in The Quarterly Journal of Economics found that access to a generative AI assistant increased customer-support productivity by 15% on average, with the largest gains for less experienced workers. A Science study of generative AI in professional writing tasks found that ChatGPT reduced time spent and improved output quality in a controlled experiment.
But task gains do not automatically become firm gains. A company can make one workflow faster while leaving approvals, handoffs, incentives, data access, and accountability untouched. That is why the NBER executives expect more tomorrow than they have seen today: they forecast AI will raise productivity by 1.4% over the next three years, raise output by 0.8%, and reduce employment by 0.7%.
The real question is whether those productivity gains will compound through redesigned work or dissipate into faster emails and prettier slide decks. Goldman Sachs Research has argued that generative AI could eventually lift productivity growth materially over a decade, while Daron Acemoglu’s macroeconomic analysis estimates much more modest total factor productivity gains over ten years.
The lesson is not that the optimists or skeptics have already won. It is that firm productivity will depend less on software access than on implementation quality.
AI creates leverage where the work is measurable, repeatable, and connected to decisions that matter. It creates noise where executives mistake experimentation for deployment.
The Jobs Story Is An Expectations Gap
The most politically explosive finding in the NBER paper is the split between executives and workers. Executives expect AI to cut employment by 0.7% across the four surveyed countries over the next three years, with U.S. executives expecting a 1.2% decline. Employees, by contrast, expect AI to raise employment at their firms by 0.5%.
That AI employment gap matters because expectations shape behavior. Executives planning for substitution will slow hiring, redesign roles, and look for automation payback. Employees expecting augmentation will invest in learning and push for tools that make them more valuable. A workplace where leaders quietly forecast job cuts while workers openly expect job growth is a workplace primed for distrust.
The broader labor market story is still unsettled. AI can reduce headcount in mature firms while creating jobs in new firms, new services, and new categories of demand. The NBER authors explicitly note that their firm sample cannot capture employment at firms that do not yet exist, which is where many technology-driven job gains historically appear.
Still, leaders should not hide behind abstraction. McKinsey’s 2025 enterprise AI survey reports that most organizations are using AI, while many remain early in scaling and capturing enterprise-level value.
Pew Research Center’s finding that 34% of U.S. adults have used ChatGPT shows public familiarity is rising, but familiarity is not the same as readiness for role redesign.
The companies that handle this transition well will be frank about where work is changing and disciplined about where people still create advantage. They will train managers to redesign jobs, not merely license tools. They will build governance for AI decision-making before bad incentives turn automation into a cost-cutting reflex.
AI is already common enough to stop being impressive. The strategic prize now is conversion. Firms need to convert trials into workflows, workflows into measurable output, and measurable output into trust with employees who can see exactly how the bargain is changing.
The delayed payoff is not proof that AI is overhyped. It is proof that technology alone does not reorganize a company. Leaders do.
