Generative AI tools are producing real, measurable gains for individual workers, but the broader economic payoff that businesses and policymakers have been banking on has yet to show up in national productivity data. Peer-reviewed research confirms that AI assistants can speed up tasks and lift output quality, especially for less-experienced employees. Yet with fewer than one in 20 U.S. firms actively using the technology as recently as early 2024, and most workers still lacking formal training, the gap between micro-level wins and a macro-level productivity boom remains wide.
Contact Centers Show What AI Can Do
The strongest evidence for AI-driven productivity gains comes from a study published in a leading economics journal that tracked a GPT-based assistance tool deployed in a real contact-center environment. The research found that the tool raised measurable worker productivity and quality, with especially large gains for less-experienced agents. Newer employees, who had the most room to improve, saw the sharpest increases, while veteran staff benefited less dramatically. This pattern, sometimes called “skill compression,” suggests that AI can narrow the performance gap between top and bottom performers on a team rather than simply making the best workers faster.
A separate preregistered experiment by Shakked Noy and Whitney Zhang reinforced that finding. Their study, covered by MIT News, showed that ChatGPT produced measurable time and quality effects on professional writing tasks. Workers who started with lower baseline performance gained the most. Taken together, these results paint a consistent picture: AI tools act as a performance floor-raiser, not just a ceiling-lifter.
Adoption Is Growing but Still Thin
If AI is so effective at the task level, why has it not yet reshaped the broader economy? Part of the answer is simple math. According to a U.S. Census Bureau working paper using high-frequency data from the Business Trends and Outlook Survey, AI use rose from 3.7% to 5.4% of firms between September 2023 and February 2024. That pace is notable, but even after the jump, roughly 19 out of 20 businesses reported no AI use at all. The same data showed that firms expected further increases, yet intentions and deployment are different things.
Survey-based research from the National Bureau of Economic Research found that work adoption of generative AI has been comparable to the personal computer relative to each technology’s first mass-market launch. That comparison is instructive in both directions: the PC eventually transformed white-collar work, but the transformation took well over a decade to register clearly in aggregate productivity statistics. Generative AI may be on a similar trajectory, moving quickly in percentage terms while still reaching only a fraction of the workforce.
Internationally, early evidence points to the same pattern. An analysis from the International Monetary Fund on AI and productivity in Europe concludes that while firms experimenting with generative tools report efficiency gains, those improvements are still too small and too concentrated to shift regional productivity trends. AI is present, but it is thinly spread and unevenly integrated into day-to-day operations.
Task-Level Tweaks, Not Organizational Overhauls
Even inside companies that have adopted AI, the technology is changing how people spend their time more than it is changing what organizations do. A randomized field experiment conducted across 66 firms and involving 7,137 knowledge workers, documented in an NBER working paper, found that generative AI shifted time allocation and work patterns at the task level. Workers used the tools to handle routine components of their jobs faster, freeing time for other activities. But the study stopped short of showing that firms had redesigned roles, restructured teams, or rethought their products around AI capabilities.
MIT economist Daron Acemoglu has offered a pointed explanation for this limitation. In his analysis of AI’s economic effects, reported by MIT Sloan, Acemoglu argued that generative AI has mostly been applied to “easy-to-learn tasks” where the connection between action and outcome is straightforward. Drafting emails, summarizing documents, and answering routine customer questions all fit that profile. The harder work of judgment, strategy, and complex coordination has largely remained untouched, which limits how much overall output can improve.
This distinction matters for anyone expecting a quick economic payoff. The organizational rewiring needed to unlock deeper gains (rethinking workflows, retraining managers, and redesigning products) has barely begun, and history suggests that such rewiring is what separates a useful tool from an economic revolution. The lag between adopting a technology and reorganizing around it helps explain why individual productivity studies look so promising while aggregate statistics remain muted.
The Training Gap Holding Back Results
A less discussed bottleneck is the gap between tool availability and worker readiness. Research reported by the London School of Economics found that 68% of employees had received no AI training in the past 12 months. That figure is striking given the volume of corporate enthusiasm around AI adoption. Companies are purchasing licenses and announcing AI strategies, but most of their workers have not been shown how to use the tools effectively.
The training deficit creates a compounding problem. The contact-center and writing-task studies both showed that less-experienced workers gained the most from AI assistance. If those workers never receive training, the skill-compression benefit disappears and organizations risk entrenching inequality instead of reducing it. Well-resourced teams and self-motivated early adopters pull ahead, while the majority of employees continue working much as they did before, leaving most of the potential productivity gains unrealized.
Evidence from broader ecosystem tracking backs up this uneven picture. The annual AI Index compiled by Stanford’s Human-Centered AI Institute notes in its 2025 report that while AI capabilities and investment have surged, measures of workforce preparedness and organizational change lag behind. Training programs, standards for responsible deployment, and practical guidance for managers are developing more slowly than the underlying models.
From Micro Gains to Macro Impact
For now, generative AI’s economic story is one of strong local effects and weak aggregate signals. In controlled settings, the technology reliably speeds up routine tasks and lifts the floor under less-experienced workers. In the real world, however, adoption is still limited, organizational redesign is tentative, and most employees have not been trained to use the tools at all. Until those constraints ease, national productivity statistics are unlikely to reflect the promise visible in individual experiments.
That does not mean the promise is illusory. Instead, the research suggests a roadmap for moving from scattered micro-level wins to a broader productivity payoff. Firms that want to get ahead will need to treat AI not just as a software purchase but as a catalyst for redesigning workflows and roles. Policymakers hoping for growth tailwinds will have to invest in training and diffusion, not just in model development. And workers, especially those early in their careers, stand to gain the most from learning how to collaborate with these tools rather than compete against them.
The history of general-purpose technologies, from electricity to personal computers, shows that the economic revolution arrives only after organizations change how they work. Generative AI appears to be following the same script. The tools are here, and their benefits in narrow settings are well documented. The question now is how quickly businesses and institutions can close the gaps in adoption, training, and organizational design so that those benefits show up not just in case studies, but in the productivity statistics that shape living standards over time.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
