I recently noticed a paradox among a team of developers. With AI, engineers started writing code faster and getting answers in seconds, yet they also reported feeling more exhausted than before.
AI hasn’t actually reduced the amount of work that needs to be done. Instead, it has fundamentally changed its nature. We can now run multiple tasks in parallel and perceive this as productivity. Up to a point, it is. But eventually, managing tools and constantly switching between them becomes more draining than performing the original tasks themselves. In some cases, it even slows down the process of finding a solution.
I’ve been managing developer teams for over 15 years, and I’ve spent the past year trying to understand why AI tools—designed to make work easier—sometimes have the opposite effect. Here are the causes behind this phenomenon and what we can do about it.
WHERE THE FATIGUE COMES FROM WHEN AI IS DOING PART OF THE WORK
Take a developer’s workflow as an example. In the past, when faced with a complex problem, developers would search Google, use Stack Overflow (before ChatGPT arrived), and ask colleagues for help. Each step and decision was separated by reflection time.
Now, they start using Cursor or GitHub Copilot—AI tools that suggest code in real time. The path to an answer gets shorter. But instead of searching, they’re now engaged in continuous evaluation of AI suggestions: Accept the autocomplete or reject it, rewrite the prompt or regenerate the output. Dozens of micro-decisions with no pauses between them.
Each of these carries a cognitive “cost.” Even the smallest choice demands attention and mental effort. The more decisions a person makes daily, the worse the quality of each subsequent one. This happens because of what psychologists call decision fatigue. AI has amplified this problem by dramatically increasing the number of decisions a person must make while completing a single task.
Researchers at Boston Consulting Group (BCG) surveyed nearly 1,500 U.S.-based workers. They found that 14% of people who use AI at work, needing a high amount oversight, experience “AI brain fry”—a feeling of mental fog and an inability to focus. And this has consequences: Workers experiencing it are more likely to consider changing jobs and make more mistakes.
MORE TOOLS DON’T IMPROVE PRODUCTIVITY
I’ve seen it repeatedly: Managers begin implementing AI with the same question: How can we use these tools to help the team get more done? Then they start adding AI services.
One or two AI tools genuinely do boost productivity, per BCG, but at three tools, productivity peaks. With the fourth, it drops. Each new tool means new settings, prompts, and workflows. At some point, the team spends more effort managing tools than doing the actual work.
The worker stops being the doer and becomes the one checking, comparing, and choosing. Meanwhile, people remain convinced that AI makes them faster. But according to METR data, the opposite was true: Experienced developers using AI tools actually worked more slowly—even while believing their task completion time decreased by nearly a quarter.
There’s another paradox here. Even when AI genuinely speeds up work, people don’t use that time to rest. They take on more tasks. This was discovered by researchers at UC Berkeley’s business school, who spent eight months observing employees at an American tech company to understand how AI usage affected their work habits.
At first, employees felt energized; their productivity soared. But over time, the workday quietly stretched longer—a prompt during lunch, one last query before leaving the office—while the number of breaks decreased. No one demanded they work more, but that’s exactly what happened. Later, those same workers admitted they were exhausted. The researchers called this “workload creep”—a gradual increase in workload that accumulates unnoticed until fatigue starts affecting decision quality.
SHOULD WE ABANDON AI TOOLS?
I don’t think abandoning AI is the answer. I’m convinced the problem isn’t the technology but how we use it and our goals. Here are five things that, in my experience, can help implement AI without burning out your team:
1. Start by rethinking your workflows. Before introducing any AI tool into a process, begin with a question: Which tasks require human attention, and which can be automated without sacrificing quality? The approach of “implementing AI in every process” isn’t a strategy—it’s a fast track to breaking what already works.
2. Give managers a leading role in AI adoption. In teams where the manager personally helps people learn AI tools, cognitive fatigue among workers is lower, according to the BCG study. A manager who truly understands how AI works can set a healthy pace for using these tools and prevent the team from drowning in experiments.
3. Establish rules for working with AI. The Berkeley researchers called this “AI practice”—agreements about how the team engages with AI. These might include a short pause before important decisions, sequential execution instead of multitasking, and time for discussion and collective reflection. One of our team leads, for example, encourages juniors to argue with AI more often.
4. Track cognitive load. We regularly conduct team health checks—monitoring productivity, engagement, and stress levels. But I’ve realized that’s no longer enough. In our new reality, cognitive load needs to become a separate metric. You can start with a few questions: How many AI tools is someone using, has AI simplified their work or increased its volume, and how does the employee feel at the end of the day?
5. Explain the reasons behind changes to your team. People can be skeptical of AI because of uncertainty. If a company doesn’t explain why it’s introducing new tools, employees start interpreting it themselves. By contrast, gaining an understanding that the company values balance—rather than simply wanting more output for the same cost—reduces mental fatigue by 28%, per the BCG research. This is exactly the approach I follow with my 100-person software team: transparency.
The key question isn’t “how do we use AI” but “why?” Start with the goal of freeing people from routine tasks. Improving decision quality will yield different results than measuring implementation success in tokens or lines of code.
Illia Smoliienko is chief software officer of Waites.