Researchers at Boston Consulting Group have identified a new phenomenon called “AI brain fry”, the acute mental fatigue caused by supervising AI tools.

The study, published in Harvard Business Review, shows that instead of freeing up our time, managing AI agents can create cognitive overload. People report more frequent errors, difficulty concentrating, and an increased intention to quit.

I can confirm this phenomenon from my own experience as a sensation of constant “mental noise,” as if I had dozens of tabs open simultaneously in my mind.

Contrary to the promise of focusing on strategic work, multitasking becomes the dominant trait of working with AI. I’m simultaneously juggling AI-generated work, reviews, corrections, and decisions across multiple projects. It sounds great in theory, but in practice it’s like micromanaging step by step a team of 10 people, each of whom asks me almost simultaneously what to do.

The temptation to use multiple agents concurrently is strong, and the promise of productivity is seductive. Since we’re still in the early stages of this form of human-machine collaboration, it’s worth remembering that we are human. We have cognitive capacity limits, and it’s important to recognize them before we reach burnout.

A few suggestions for you and the teams you manage to limit AI brain fry:

1. Start with a single agent, a single workflow, and don’t scale before you have a stable and predictable process. Chaotic automation only produces chaos faster. Don’t ask employees to manage more than one agent simultaneously.
2. Clearly define what you decide and what the AI decides. Without this boundary, you end up reviewing everything (essentially redoing the work manually with extra steps) and in reality, productivity drops.
3. Schedule breaks and pace your deadlines. The BCG study confirms that brain fry disappears when you take a break.
4. Invest in your own and your team’s education before giving access to AI tools. Teams that understand how the technology works have realistic expectations and are far more productive.

GenAI-powered tools still require human oversight, and oversight consumes mental energy and this is likely a transitional problem.

The tools are new, people are still learning to manage them, and it’s like handing a Ferrari to someone who just got their driver’s license.