AI Made You Fast. Now Make It Sustainable.

Earlier this year, a programmer released a tool called Gas Town that lets you run swarms of AI coding agents in parallel. Ship software at blistering speed. One user tried it and had to stop. Not because it didn't work, but because it worked so well that watching it felt overwhelming. Too many things happening, too fast, for a human brain to track.

I recognise that feeling. On any given morning I'm triaging output from my personal AI agent (Gizmo), reviewing research from a Cowork scheduled task that ran overnight, checking on build experiments, and trying to remember which of three tools I kicked something off in. And that's before I've started the work I actually sat down to do.

That feeling now has a name. Researchers at BCG surveyed nearly 1,500 workers in January and identified a phenomenon they're calling "AI brain fry": mental fatigue from the constant oversight, evaluation, and iteration that intensive AI use demands. Fourteen percent of AI-using workers reported experiencing it. They described it as fog, buzzing, a sense of having twelve browser tabs open in their head at once.

Brain fry is not burnout. That distinction matters. Burnout is emotional exhaustion and disengagement. Brain fry is cognitive: your brain literally cannot keep up with what the tools are asking of it. And here's the finding that should change how every leader thinks about AI deployment: the same study showed that AI can reduce burnout (by automating repetitive tasks, workers reported 15% lower burnout scores) while simultaneously increasing brain fry. You solve one problem and create another. The drudge work disappears, but it gets replaced by a different kind of mental load: constant monitoring, endless iteration, and the growing feeling that you're working harder to manage the tools than to solve the actual problem.

The costs are not abstract

Brain fry isn't a wellbeing issue you can file under "nice to address eventually." It's a business performance problem with hard numbers.

Workers experiencing brain fry report 33% more decision fatigue. They make 39% more major errors, the kind that affect safety, outcomes, or important decisions. And 34% of workers with brain fry are actively planning to leave their jobs, compared to 25% without it. That's a 39% increase in quit intent among your heaviest AI users, the people most companies are desperate to retain.

There's also a specific threshold where productivity tips over. Workers using up to three AI tools reported genuine gains. At four tools, perceived productivity dropped below where it started with two. The multitasking trap, back with a vengeance. Meta is now counting AI-generated lines of code as a performance metric for engineers. When organisations incentivise volume of AI use rather than quality of outcomes, they're systematically manufacturing brain fry.

The messaging problem

BCG uncovered something that every leader deploying AI needs to hear: when employees feel their organisation expects them to accomplish more work because of AI, mental fatigue scores jump 12%. And that expectation doesn't need to be stated explicitly. References to "productivity gains" and "efficiency" are enough. Workers hear the subtext: you're now expected to do more.

This is the invisible tax on every AI rollout that leads with efficiency. Even if that's not what you meant, that's what lands. I've sat in enough rooms where leadership announces AI-driven productivity improvements and watched the body language shift. People don't hear "we're making your job easier." They hear "we expect more from you now."

Meanwhile, Anthropic just published what may be the largest qualitative study ever conducted: 81,000 users across 159 countries, asked what they actually want from AI. Professional excellence topped the list, but when researchers pushed on why, the answers drifted. More time with family. Mental breathing room. A better life. A third of respondents weren't chasing productivity at all. They wanted relief from the relentless pace.

The gap between what organisations are selling (more output, faster) and what people actually need (sustainable capacity that translates into a life worth living) is exactly where brain fry grows.

What sustainable AI productivity looks like

The good news: brain fry isn't inevitable. The BCG data points to specific conditions that reduce it. None of them are complicated.

Limit the tools you use. Three tools is the ceiling before productivity reverses. Before adding a new AI tool, decide what you're retiring. Depth with fewer tools beats shallow use across many.

Define "done" before you start. The endless iteration loop is the core driver of cognitive overload. Without AI, a marketer writes one campaign brief. With AI, they generate twenty variants and then have to evaluate all twenty. Marketing roles reported 26% brain fry rates, the highest of any function. I notice this in my own building too: when is an experiment done? Does it have more legs, or am I just iterating because the tool makes it frictionless to keep going? Set quality thresholds upfront. "Good enough" is a decision you make before you prompt, not after your tenth revision.

Shift metrics from activity to impact. If you're measuring token consumption, lines of AI-generated code, or adoption rates, you're incentivising the wrong behaviour. Measure outcomes. And when a worker automates something clever, don't immediately backfill their time with more work.

Make AI a team practice, not a solo sport. Brain fry drops significantly when teams integrate AI into shared workflows rather than leaving each person to figure it out alone. When one person prompts, another evaluates, and another integrates, the cognitive load distributes. Teams with organised AI integration reported significantly less mental strain than those where everyone was on their own.

Be the manager who engages. Workers whose managers actively discuss AI use with them had 15% lower mental fatigue scores. Workers whose managers expected them to figure it out alone had measurably higher scores. BCG calls this the "AI orphan tax." Twenty minutes of genuine conversation about how your team is using AI, what's working, and what's draining them is worth more than another tool demo.

Say the quiet part out loud. If your organisation is deploying AI, be explicit about workload expectations. Will output targets increase? Will headcount change? The worst thing you can do is stay silent and let people fill the gap with anxiety. Employees who felt their organisation valued work-life balance had 28% lower mental fatigue. That signal matters enormously.

The only productivity that compounds

The companies that win the next phase of AI won't be the ones with the highest adoption rates. They'll be the ones whose people can sustain their performance over years, not just quarters.

Sustainable AI productivity isn't the soft version of productivity. It's the only version that doesn't collapse under its own weight.

Do this Monday morning: count your AI tools. If you're above three, consolidate. Set one "done" threshold for your most common AI-assisted task. And if you manage people, book 20 minutes this week to ask your team not how much AI they're using, but how it's making them feel.

The Atomic Builder is written by Faisal Shariff and powered by Atomic Theory Consulting Ltd — helping organisations put AI transformation into practice.