The productivity paradox: why AI tools are making workers slower

The burnout crisis and the productivity push

In the past year, 81% of C-suite leaders increased demands on their employees, asking them to do more with the help of AI, expand their skill sets, and work longer hours. The result, according to a new survey from the Upwork Research Institute, is that 71% of full-time employees now report being burned out, and 65% say they are struggling to keep up with their employer's productivity expectations. Gen Z workers are hit hardest, with 83% burned out, followed by 73% of Millennials.

The research, conducted in April and May 2024, surveyed 2,500 workers across the U.S., UK, Australia, and Canada, including C-suite executives, full-time employees, and freelancers. The data shows a workforce stretched thin, even as leaders push for greater efficiency through AI adoption.

The AI productivity paradox

Executives are overwhelmingly optimistic about AI. Ninety-six percent of C-suite leaders expect AI tools to increase their company's overall productivity. Already, 39% of companies mandate AI use, with another 46% encouraging it. But workers report a different picture.

Among employees using AI, 77% say the tools have decreased their productivity and added to their workload. They report spending more time reviewing AI-generated content (39%), learning to use the tools (23%), and simply being asked to do more work (21%). Forty-seven percent of workers say they have no idea how to achieve the productivity gains their employers expect.

This is not the first time a promising technology has failed to deliver immediate productivity returns. The study refers to the 'productivity paradox' observed by economists like Robert Solow, who famously quipped: 'You can see the computer age everywhere but in the productivity statistics.' Generative AI may be repeating that pattern, echoing the way benchmark results can mislead in AI evaluation itself, as the MMLU contamination case shows.

The leadership perception gap

Leaders and employees describe their company differently, according to the survey. While 84% of C-suite leaders say their companies value employee well-being over productivity, only 60% of full-time employees agree. Leaders also overestimate their workforce's AI readiness: 37% say their employees are 'highly' skilled with AI tools, but just 17% of employees report that level of comfort.

Only 26% of companies have AI training programs in place, and just 13% report a well-implemented AI strategy. Most AI use is emerging bottom-up, with workers leading the charge. But when leadership pushes adoption from the top down without changing workflows, it creates what the researchers call 'productivity strain'. Claims of AI readiness in the industry often don't hold up under real-world testing, as a study on harness evolution found.

Why mandating AI backfires

Mandating AI use without proper support appears to worsen burnout. Workers whose companies require AI are more likely to feel overwhelmed: 40% say their company is asking too much of them regarding AI. The study finds that employees who perceive their company to value productivity over well-being are more likely to feel overwhelmed by their workload (73% vs. 56%).

Nearly half of workers using AI say they have no idea how to achieve the expected productivity gains. The combination of top-down mandates and unclear expectations creates a cycle where AI becomes another task rather than a tool for efficiency. A similar pattern appears in AI model orchestration: relying on one large model can underperform a targeted swarm of specialists, as the Claude vs Fugu analysis shows.

Pathways to balance: freelancers and co-created metrics

The research proposes three steps for leaders who want to break the cycle. First, leverage nontraditional talent. Nearly half of freelancers (48%) say they are AI-skilled, compared with far fewer full-time employees. C-suite executives who hire freelancers for AI projects report at least doubling organizational agility, innovation, and efficiency. Eighty percent of leaders who use freelance talent say it is essential to their business.

Second, co-create measures of productivity. Fifty-four percent of employees say their company does not have an accurate picture of their productivity, and 74% believe the approach needs an overhaul. When workers have a say in how they are evaluated, satisfaction and productivity both rise. The study reports that 81% of employees would be more satisfied and 76% more productive if they had more input.

Third, build fluency in the language of skills. Only 40% of executives claim high awareness of the AI skills in their workforce. Without that map, AI initiatives are flying blind. The study argues that skills-based approaches, rather than job-title-based ones, are essential for effective AI deployment. The value of specialization is evident even in model design, where splitting capabilities across multiple experts has delivered top performance, as the Macaron-V1-Venti model illustrates.

The Upwork study suggests that the path to AI-driven productivity is not through mandates and top-down pressure, but through rethinking how work is organized. As the research puts it: 'Doing more with less, ignoring alternative talent pools, and sticking with top-down productivity measurement simply won't work in the era of AI.'