Why do AI's most productive users burn out the fastest?

The Upwork Research Institute, working with Workplace Intelligence, surveyed 2,500 global workers (1,250 C-suite executives, 625 full-time employees, 625 freelancers, fielded by Walr between March 25 and April 9, 2025) and finds that "employees using AI report a 40% boost in productivity on average," but "workers who report the highest productivity gains due to AI are also the most burned out: 88% say they are experiencing burnout, and they are twice as likely to consider quitting" than less-productive AI users. The same high-performing group reports weaker human ties: "more than two-thirds of high-performing AI users say they trust AI more than their coworkers," 64% say they "have a better relationship with AI than with human colleagues," and a majority find AI "more polite and empathetic than their teammates." Nearly half say "please" and "thank you" on every request, and 87% phrase requests "as if to a human coworker at least some of the time."

The report frames this as a single mechanism rather than two separate findings: Dr. Kelly Monahan calls it a "productivity paradox," arguing that "traditional work systems... reward output with AI, but overlook the human relationships behind that work." It supplies a second, related gap behind the burnout — 62% of high-performing AI users "don't understand how their daily use of AI aligns with company goals" — and reads the combination (unsupported top performers, degraded human ties) as what converts proficiency into flight risk, rather than treating burnout as an independent side effect of heavier workload.

The strategy disconnect half of this claim parallels Why do ready workers struggle to transform their work?, where ready individual users outpace organizational incentive and leadership design — Microsoft's gap sits between worker capability and management support, Upwork's between daily AI use and stated company strategy, and both put the missing piece above the individual. The burnout finding complicates Does generative AI actually save workers time or intensify it?: Ye and Ranganathan locate intensification in self-driven scope creep and dissolved pauses, while Upwork's number suggests the workers intensifying the most (the highest-productivity group) are also the ones burning out, a correlation consistent with but not proof of that mechanism. The "trust AI more than coworkers" finding sits in tension with Why do workers hide productivity gains from AI use?: Anthropic's subjects conceal AI use from colleagues out of stigma while privately valuing it, whereas Upwork's high performers openly prefer AI's company to their coworkers' — both describe a private AI relationship outrunning public workplace norms, approached from opposite directions (concealment versus stated preference).

Every figure is self-reported on a cross-sectional survey commissioned by a freelance marketplace with a commercial stake in AI-skilled talent (Upwork's own framing notes "nearly 300%" growth in searches for AI-agent freelancers), so the 40% productivity figure, the 88% burnout figure, and the "twice as likely to quit" comparison are all perceptions elicited in the same instrument, not independently measured output, attrition, or AI-use logs. The report does not establish that AI productivity gains cause burnout or that trusting AI over coworkers causes disconnection from strategy — only that the two patterns co-occur among workers who report doing the most with AI. The most defensible implication is narrower than "AI relationships at work": that whatever makes some workers highly productive with AI, in this sample, tends to arrive bundled with weaker organizational support and weaker human ties, which is a reason to test support interventions, not evidence that AI itself degrades coworker trust.