Harvard Business Review recently published “When Using AI Leads to Brain Fry”, surveying 1,488 full-time workers. Finding: 14% of AI users experience what they call “AI brain fry.” I read it and thought — this is exactly what I’ve been living through. In my last post, I argued that the bottleneck eventually lands on the human brain. This HBR study confirms it’s not just me. It’s structural.

What AI brain fry looks like: mental exhaustion, a buzzing sensation, brain fog, difficulty focusing, slower decision-making, headaches, and the constant urge to stop and reset. Imagine fifty browser tabs open inside your skull. The primary cause: high-intensity AI supervision — monitoring multiple tool outputs simultaneously.

Vibe coding is, by definition, a high-cognitive task. Burnout isn’t a risk — it’s a guarantee. The question is how you manage it.

Hardware fixes. Since the bottleneck is biological, attack it biologically. Higher quality sleep and exercise — weight training plus cardio — are the highest-leverage resets. I also supplement daily: zinc and magnesium for nerve recovery and sleep, curcumin for inflammation, fish oil (Omega-3) as essential brain nutrition, and creatine for a slight energy boost. All safe, all evidence-backed. I’ve been on this stack since my gym days — stable, no issues. Diet matters too: cut the salt and grease. Mediterranean diet works best.

Software fixes. Reduce the number of decisions you make during development. The ideal: AI that can one-shot the entire task. The debugging loop is what really drains you. Invest in making your AI more capable — better scaffolding, better prompts, whatever it takes — and you reduce cognitive load at the root.

Don’t open too many sessions. I used to run nine terminals. Unnecessary. Two to three active development sessions is enough to keep most people fully loaded. Beyond three, you’re multi-tasking, and humans are bad at multi-tasking.

This is especially true at night. By evening, your brain’s context window is running low — decision quality drops off a cliff. You need sleep to compact. From my own data: coding at night produces significantly more bugs. Mornings don’t. Some of our cognitive mechanisms are remarkably similar to how LLMs work. Fascinating.

Pushing the ceiling of vibe coding requires biohacking.