AI brain fry is real. Here’s what’s causing it | Julie Bedard

AI “brain fry” is different from burnout: Julie Bedard describes it as acute mental strain from overseeing AI tools, which can bring mental fog, slower decisions, and more errors. Workers who reported brain fry also reported 39% more major errors and were 39% more likely to say they intended to leave. These are associations in the reported data, not proof that AI oversight caused either outcome.

The strain can arise when AI is layered onto existing work without removing tasks or redesigning workflows across the team. Oversight itself can add cognitively demanding work, particularly when employees must check AI output while still meeting their original responsibilities. In Bedard’s data, managers who answered employees’ AI questions had teams with mental-fatigue scores 15% lower. One manager described her team producing 10 times more output 10 times faster, while her own job felt 20 times harder and lonelier; the work was sometimes better and sometimes worse.

Bedard says more than half of jobs will be dramatically reshaped in the coming years, but individual productivity gains do not automatically scale into better team performance. A person may redesign their own tasks around AI, yet broader gains require teams to reconsider who does what, how work moves, and who checks the results. She argues that leaders need to set expectations for attention and accountability and preserve the support people need to learn their roles.

An experiment illustrates why accountability needs explicit design. Managers reviewed the same document with the same errors, but they attributed less personal accountability when the work was labeled as coming from an “AI employee” than when it came from an AI tool or a human employee. Among managers familiar with AI agents on org charts, accountability fell by 9 percentage points, they found 18% fewer errors, and escalations rose 44%. Bedard says organizations should give AI agents human overseers with clear decision rights and escalation paths, and train employees to review AI-generated content before assigning agents a formal place on the org chart.

AI can take on some entry-level drafting, analysis, research, and first-cut work, but those tasks have traditionally helped junior employees develop judgment. Bedard cautions that a junior employee may produce good work on a stretch assignment with AI without building that judgment if expert support is missing. She sees a way to retain the learning while using AI: make jobs more rotational and project-based, pair junior staff with senior experts, and draw on junior employees’ AI fluency. The trade-off is that automating early-career tasks can improve immediate output while weakening the path by which future experts gain experience.

Bedard says managers’ roles and competencies need to change too, emphasizing coaching, AI upskilling, and team-based workflow redesign. Companies getting the most from AI were 12 times more likely to have executives and leadership teams engaged with AI, 4 times more likely to dedicate time and structure to AI learning, and 2.5 times more likely to reshape workflows. She also says anxiety, shame, and fear of missing out can reflect being inside a work transition; leaders should name the changes and actively work through them rather than leaving employees to interpret the strain alone.