A new report from the Federal Reserve Bank of St. Louis shows that output in U.S. businesses is trending higher, even though head count has barely moved. A few years ago you might have blamed pent-up demand or a lucky sales run. In late 2025, the more honest explanation is that a growing share of your team has a chatbot open in the background.

The St. Louis Fed’s national U.S. adoption tracker, built on its Real-Time Population Survey, shows generative AI use jumping 10% in a single year. The new analysis of adoption and productivity argues those extra minutes are starting to show up in macro data.

AI used by majority of working-age Americans

Gen AI use is already a majority behavior for working-age Americans. The latest Real-Time Population Survey data show that by August 2025, 54.6% of adults aged 18–64 had used gen AI, up from 44.6% a year earlier, with work use rising from 33.3 to 37.4%, and nonwork use from 36.0 to 48.7%. Three years after launch, this adoption rate is far ahead of personal computers and the early commercial internet at comparable points in their rollout.

An earlier working paper on adoption using the same survey already found that nearly 40% of adults were using gen AI by late 2024, with between 1% and 5% of all work hours assisted by the technology. In other words, what looked like a wave of experimentation has hardened into routine use. For managers, that means your workforce is no longer waiting for a formal AI strategy. They are already automating pieces of their day, even if your policies and metrics haven’t caught up.

The picture is global, not just American. A 2024 global employee survey of more than 13,000 workers in 15 countries found that about half of employees using gen AI saved at least five hours a week, and nearly two-thirds of leaders said they were starting to redesign their organizations around it. Microsoft and LinkedIn’s 2024 Work Trend Index report similarly reports that 75% of knowledge workers worldwide are already using AI, with almost half starting within the previous six months, and many doing so ahead of any official guidance.

Shadow use is now a structural feature of the workplace. A recent study of “bring your own AI” behavior based on payroll and survey data finds that nearly half of U.S. workers use AI tools without telling their employer, and roughly two-thirds of those users pay for the tools out of pocket. The combination of high adoption and low formal oversight means leaders who rely only on sanctioned tool metrics are likely underestimating how deeply AI is already woven into everyday work.

AI productivity gains

The strongest evidence for productivity gains comes from narrow tasks, and it’s no longer limited to lab settings. The St. Louis Fed’s work productivity analysis estimates that among workers who used gen AI in the previous week, average time savings reached 5.4% of their work hours, with 20.5% of these users saving four or more hours per week. When you include nonusers, that still translates into 1.4% of total hours saved throughout the workforce.

Randomized experiments reinforce these self-reports. In a large customer support experiment with 5,000 agents, access to a gen AI assistant increased issues resolved per hour by 14% on average, with the biggest gains for novice workers and minimal gains for seasoned experts. In software development, a trio of GitHub Copilot field experiments at Microsoft, Accenture, and a Fortune 100 manufacturer found that developers with access to the tool increased weekly pull requests by about 26%, again with outsized benefits for junior engineers. A separate professional writing experiment shows that giving knowledge workers access to ChatGPT cut completion times by roughly 40% and improved quality scores by double digits.

The Real-Time Population Survey team at the St. Louis Fed has now connected these microlevel gains to the broader economy. Pooling survey waves from early 2025, they estimate that self-reported time savings from gen AI correspond to 1.6% of all U.S. work hours, implying up to a 1.3% boost to labor productivity since ChatGPT’s release when fed into a standard production model. That estimate lines up with official statistics: Labor productivity in the U.S. nonfarm business sector grew at an annualized 2.16% from late 2022 through mid-2025, compared with 1.43% per year in the 2015–2019 period cited in the same analysis.

Not all of that gap comes from chatbots, of course. Some saved time turns into on-the-job leisure rather than extra output, a point emphasized in both the St. Louis Fed work and an ITIF commentary on time savings. Yet even if only part of the reported 5% to 25% task-level improvements are captured as throughput, the cumulative effect on project timelines, service quality, and innovation pipelines is significant. For professionals managing complex portfolios, that translates into extra cycles for client work, experimentation, and strategic planning that rarely fit into traditional schedules.

The next phase is less about whether gen AI works and more about how firms convert scattered time savings into durable performance.

The next phase is less about whether gen AI works and more about how firms convert scattered time savings into durable performance. Global modeling from McKinsey estimates that recent advances in gen AI have raised the share of work hours that are technically automatable from about 50% to as much as 60–70% and could add 0.1 to 0.6% to annual productivity growth between 2023 and 2040, within a broader automation range of 0.5 to 3.4%. Those gains only materialize if organizations actually redesign workflows so that freed-up hours are redeployed into high-value activities rather than drowned in meetings and email.

The St. Louis Fed’s new analysis offers an early stress test. By correlating industry-level gen AI time savings with detrended productivity growth, the authors find that industries reporting 1% higher time savings saw, on average, 2.7% faster productivity growth relative to their prepandemic trend, with a correlation of 0.32 across sectors in their industry-level correlation study. It’s explicit that this pattern is not proof of causality, but is exactly the sort of relationship you would expect if AI-assisted work was beginning to matter in the aggregate.

Still, firm adoption lags behind worker behavior. Even among adopters, use often remains confined to marketing automation and analytics pilots rather than end-to-end process redesign. That gap between individual experimentation and organizational commitment shows up clearly in the Work Trend Index, where high employee use coexists with the finding that 60% of leaders say their organization lacks a clear AI plan.

For executives, the implication is blunt. The technology has already crossed the adoption threshold. The differentiator now is whether your organization treats gen AI as a sanctioned part of core workflows. That means mapping tasks where workers already use AI informally, standardizing prompts and guardrails, investing in targeted training, and tying AI-assisted work to performance metrics rather than leaving it in the shadows. Companies that take this operational route are more likely to convert scattered time savings into measurable gains in throughput, quality, and innovation. Those that don’t may still see happier employees, but they will leave much of the productivity upside on the table.