For three years, the sales pitch has been identical. We are told that AI is a “force multiplier” that handles the grunt work so we can go home early. But researchers spent eight months inside a tech company watching this exact scenario play out, and they found something different. Instead of liberation, they found a staff that was working through lunch, logging on late at night, and slowly burning out.
Key Takeaways
- AI adoption at a 200-person firm increased work hours and burnout over eight months.
- Developers using AI took 19% longer on tasks while believing they were 20% faster.
- AI adoption across thousands of workplaces resulted in a 3% gain in time savings.
The researchers tracked a 200-person firm where workers genuinely embraced these new tools. Management did not demand higher targets. No one cracked a whip. The pressure came from the tools themselves. Because tasks felt easier to start, employees took on more of them.
Their to-do lists expanded to fill the time AI supposedly freed up. One engineer noted that you expect to work less, but you end up working the same amount or more. The study describes a workplace where efficiency bled into personal time, creating a cycle of fatigue rather than freedom.
The big deal
This challenges the core promise of the current tech boom. We assume productivity gains automatically translate to leisure or efficiency. This data suggests the opposite: efficiency creates a vacuum that we immediately fill with more work. The study found that organizational expectations for speed rose alongside the tools, making it harder for anyone to step away.
It also highlights a disconnect between perception and reality. A separate trial found developers using AI believed they were 20% faster, but actually took 19% longer to finish tasks. We feel faster, but we might just be spinning our wheels at higher speeds. If these tools increase stress without delivering meaningful time savings, the business case for them becomes much thinner.
How it works
The mechanism is simple: when a task becomes cheaper to do, we do it more often. Think of it like widening a highway to fix traffic. You add lanes to reduce congestion, but the extra space just invites more cars, and soon the road is just as jammed as before. AI cleared the lanes at this company, so employees immediately filled them with more tasks until they were back in gridlock.
The catch
The primary downside is the human toll. The study explicitly links this cycle to burnout and a sense that work is impossible to escape. The tools increase the volume of output without necessarily increasing the value, leaving workers exhausted.
There is also a reliability issue. In some cases, using AI actually slows people down. That separate trial with developers showed that while they felt faster, the cleanup and verification required for AI code meant the total time to completion went up, not down. The tools can create a mirage of speed while actually adding drag to the process.
What now?
Companies need to decide if “more” is actually the goal. If you manage a team, you may need to cap the number of active projects explicitly to prevent this creep. Without guardrails, the work expands until the people break. Watch to see if companies start measuring “hours saved” rather than just “output generated” in their next quarterly reports.
