The Load We’re Not Measuring
AI promised to do the work. For a lot of us, it has: the output’s faster and the capability’s broader. But, alongside the productivity gains, something else arrived that doesn’t show up in efficiency metrics.
Today, we ask: is AI intensifying work rather than reducing it?
Yes, according to an eight-month study by Berkeley researchers. In short, “the researchers found that generative AI didn’t free up time, it expanded what workers felt capable of, and willing, to take on.”
We’re in the midst of a massive divergence between output speed and biological stability as automated tools reduce the time required for individual tasks whilst exponentially increasing the cumulative load on the operator.
The AI Productivity Paradox
Upwork, a freelancer platform, in 2024 found that 77% of employees reported AI had increased their workload. In their 2025 report, having interviewed 1250 people (from C-Suite executives all the way down), they determined AI led to a 40% boost in productivity.
In tandem with that boost, there’s a serious, unignorable biological cost attached.
- 88% of the most productive AI users report experiencing high levels of burnout.
- High performers are twice as likely to consider quitting compared to less productive peers.
An astounding figure that aligns the most productive personnel with the most punishing returns. And beyond this their responses highlight an emerging social shift:
- Over two-thirds of high performing AI users say they trust AI responses more than their coworkers.
- 64% say they have a better relationship with the software than their human colleagues.
- 85% thought they were more polite to the AI than people around them.
The Social Cost of Automation Fatigue
So, AI is giving us more work but making us more productive, whilst increasing burnout dramatically and reshaping the way we work as well as the way we interact with the world.
This shift in the nuts and bolts of work is dramatic, impacting our nervous system and, in turn, our capacity. Until now we completed most tasks chronologically, starting with nothing and finishing with a solution. Throughout we navigated problems and solutions, took breaks and built regulation into the process.
Now, that’s all out the window. We bark a prompt and an answer appears out of thin air in seconds. A twelve-point piece of feedback on a legal document teleports in before we’ve swigged our coffee.
This change from fact gathering > drafting > output (pre AI) to instant output > intense auditing > verification (with AI) is the difference between a process where our system can track the build, returning to baseline once the task is complete, and a system that requires constant vigilance.
Mental Hangover and Cognitive Load
The more we use AI, the more likely we are to experience “AI Brain Fry”, as coined by the Boston Consulting Group earlier this year. It’s a term that describes “mental fatigue from excessive use or oversight of AI tools beyond one’s cognitive capacity.” A situation many of us can relate to, where an “an “intensive back-and-forth with the tools, followed by an inability to think clearly, like a mental hangover, comprised of difficulty focusing, slower decision-making [...] requiring several to physically step away from their computer to ‘reset.’”
The research found that when AI is “used to replace routine or repetitive tasks, burnout scores–but not mental fatigue scores–are lower.” But nobody uses AI in a vacuum. People’s roles that required these repetitive tasks are being phased out, with that workload being passed onto their colleagues in addition to their existing jobs, as opposed to just disappearing.
Whether internally through corporate work structures or individually in our own projects, we’re now able to do work that historically belonged to specialists. Everyone can design, write or code now. The prompt is simply “act as <job title> and do <any task>” and our robotic genie grants us our wish. We don’t need Jeff to open Photoshop, or Sinita to draft a blog post.
Workload Creep and the Ever-Moving Baseline
This is understood as “workload creep”, where the capacity of the operator expands as the barriers to entry continue to drop. More productivity means more output, more checking, more culpability for mistakes and more expectation.
The productivity gains are a hockey stick at the beginning, but then that level of output becomes our baseline. In the short term it’s great, in the long-term, according to the same Berkeley study we started with, it’s debilitating.
“In micro moments of prompting [...] people talked about […] a sense of expanded capability. But when they stepped back and reflected on their broader work experience […] they described feeling busier, more stretched. The contrast suggests that intensification can feel positive in the short bursts, while the cumulative effect creates strain over time.”
Psychologist Lisanne Bainbridge documented this dynamic in her seminal 1983 paper, Ironies of Automation. Bainbridge’s research states that the more advanced an automated system becomes, the more demanding the human supervisor's role becomes. She knew this long before Claude: automation leaves the human with the most difficult task: monitoring for rare, unpredictable errors.
Because the system appears highly capable, the penalty for a lapse in attention is incredibly high. The operator does not expend capacity on gathering facts. Instead, capacity is consumed by considering plausibility, redrafting word salad and verifying truth.
Or, as per Francesco Bonacci, founder of Cua AI, “I end each day exhausted–not from the work itself, but from managing the work.”
Where the Physiology Leads
The nervous system governs capacity. It sets the limits for what we can endure. AI tools are uniquely powerful in the pursuit of productivity, but they haven’t changed the biology of the human system that uses it.
The regulation we lose from stretching ourselves too thin with AI is crucial infrastructure we need to function properly. There’s no question whether the software is efficient or not, but what we need to ask is whether the human system operating it can use it and still return to baseline. After all, that’s the only metric worth measuring.
Leave a comment