What AI is actually doing to your brain: understanding brain fry

Executive overview

Intensive AI oversight — managing multiple agents, error-correcting, prompt-directing — creates a distinct cognitive strain called brain fry. It differs from burnout: it doesn't cause it, and the interventions that reduce burnout don't necessarily help here.

Research by Gabriella Rosen Kellerman at BCG found that productivity peaks when using two to three AI tools simultaneously, then drops. Most organisations are missing the 70% of AI adoption that matters: the people side.

The most sophisticated AI users aren't the ones with the most tabs open — they're the ones who know their own cognitive limits.

What brain fry is and how it differs from burnout

• Brain fry: mental fatigue from excessive AI oversight beyond one's cognitive capacity
• Burnout and brain fry are independent — brain fry does not increase or decrease burnout
• Using AI to replace repetitive tasks does decrease burnout; intensive multi-agent oversight does not
• Marketing and operational/technical roles show the highest brain fry rates
• Task-switching between browser tabs and tools is a core symptom trigger

The three-tool productivity cliff

• Productivity rises from one to two to three simultaneous tools, then falls beyond three
• Most sophisticated users self-select to around three tools — not the maximum they could handle
• When tools have processing delays, switching between them in rotation can work; faster tools remove that buffer
• Individual cognitive capacity varies — two or even one tool may be optimal for some people
• Data is timestamped to January 2026; optimal numbers may shift as tools evolve

What managers do that makes it worse or better

• Managers answering employee AI questions predicted 15% less mental fatigue in their teams
• Framing AI adoption as a solo mission ("learn it yourself") predicted 5% more mental fatigue
• The most protective manager behaviour is presence, support, and learning together — not AI-specific tips
• Competitive dynamics within teams (who uses AI most) increase strain; collaborative learning reduces it
• Team-level practices — shared trainings, peer teaching, embedded tools in team workflows — help

What organisational messaging does

• Messages emphasising work-life balance had the strongest effect on reducing burnout and mental fatigue
• Employees who felt AI would dramatically increase their workload were more vulnerable to mental fatigue
• The goal is not to deny heavier work is coming, but to avoid signalling it as the dominant story
• Emphasise creativity, new capabilities, and time freed for connection — not volume increases
• Counting tokens (as Meta reportedly did) measures activity, not value; better metrics focus on output quality and rate of experimentation

The 10-20-70 rule for AI adoption

• 10% of what determines successful adoption: the algorithms
• 20%: the data and tools
• 70%: processes and people — the part most organisations under-invest in
• High agency and high optimism predict who will get the most from AI tools
• Thoughtful design of how work flows, who uses which tools, and at what scale reduces change burden

Building self-awareness around cognitive limits

• Treat brain fry like learning your breath on a run — notice when it's coming and pause before you need to
• Take genuine digital breaks: no internet, no email; audio-only calls help
• At least one 24-hour digital detox per week is Kellerman's personal practice
• Regularly check in on how your brain is feeling, not just task completion
• Brain fry can feel engaging in the moment; the goal isn't to avoid it but to pace sustainably

What recovery and healthy AI use looks like

• Automate genuinely disliked repetitive tasks first — this is where the burnout reduction actually lives
• Use freed time for connection, creativity, and depth of interaction — not more output volume
• A healthy relationship with AI is characterised by optimism and a strong sense of personal agency
• Self-compassion and keeping perspective (not productivity metrics) are the primary resilience drivers