Your Thinking Muscle Collapses in 10 Minutes · Issue #153

Opening

Dear reader, did you expect AI to lighten your workload?

ActivTrak analyzed 443 million hours of digital activity from more than 160,000 workers, and the results were the exact opposite. After adopting AI, time spent on email and messaging more than doubled, and use of work software rose 94%. Meanwhile, focused work time dropped by 23 minutes. Weekend work increased by more than 40%. AI didn’t reduce work—it intensified its density and pace. Researchers have started calling this state “AI brain fry.”

But this is only the beginning. There’s something even more worrying. To cut to the conclusion: what AI is taking from us isn’t our jobs so much as our capacity to think.

🧠 Ten Minutes Is All It Takes: The Collapse of Cognitive Muscle

A joint research team from Carnegie Mellon, Oxford, MIT, and UCLA ran a randomized controlled trial with 1,222 participants this April. Participants solved math and reading comprehension problems with AI assistance, then had the AI taken away without warning (a “bait-and-switch” design).

The results were striking. When people who’d had just 10 to 15 minutes of AI help then solved the same type of problem without AI, their accuracy dropped from 0.73 to 0.57. Even more alarming was the give-up rate: the group that never used AI had an abandonment rate of 11%, while the group that had AI taken away rose to 20%. Just 10 minutes is enough to erode the will to solve a problem at all.

The team likened this to the “boiling frog effect.” Each individual instance of AI help seems trivial, but the cumulative effect eventually reaches a point that’s hard to reverse. They warned that people with fewer learning resources are especially vulnerable to this cumulative effect.

Nataliya Kosmyna’s team at the MIT Media Lab reached a similar conclusion. They had 54 participants write essays using ChatGPT while measuring their brainwaves (EEG), and found that the AI-assisted group’s brain connectivity¹ was markedly lower than the group that wrote without any tools. This phenomenon, which the team called “cognitive debt,” worsened as sessions repeated. While AI was doing the thinking, the brain was quietly powering down.

There was an interesting twist, though. In the fourth session, when the group that had been writing without any tools used ChatGPT for the first time, their brain connectivity actually increased. For people who had already built a foundation of independent thinking, AI functioned as a complement. Sequence matters. Starting with AI before building that foundation, versus layering AI on top of an already-built foundation, produces opposite outcomes.

Dr. Kosmyna explained why she released a 206-page paper before formal peer review: “I was afraid that in six to eight months, some policymaker might decide to ‘build an AI kindergarten.’ Developing brains are at the greatest risk.”

The most chilling case came from medicine. A study conducted at four endoscopy centers in Poland found that after using AI-assisted colonoscopy, when doctors then examined patients without AI, their detection rate for precancerous lesions fell from 28.4% to 22.4%—a drop of 6 percentage points, or 20% in relative terms. Skilled physicians’ diagnostic ability had degraded because of AI. The study was published in The Lancet Gastroenterology & Hepatology and is regarded as the first empirical demonstration of AI-induced skill decay among medical experts.

⚡ The Trap Called Optimization

To understand the roots of this phenomenon, we need to examine a philosophy that Silicon Valley worships.

David Brooks nailed it in his latest column for The Atlantic. Traditional education and personal growth are built on a cultivation mindset: people become resilient by enduring hard tasks, failing, and trying again. The tech industry, by contrast, is built on an optimization mindset, whose goal is to eliminate all friction and extract output as efficiently as possible.

Amit Singhal, Google Search’s former head, said in a 2013 Guardian interview that his team was “obsessively focused on eliminating every friction point between the user and the information they want.” In a world where eliminating friction is a virtue, the cultivation mindset loses its footing. The ActivTrak data shows this tension too: people who spent 7-10% of their total work time on AI tools were the most productive, but only 3% of workers fell into that range. Most people used AI either too little or leaned on it too much.

The problem is that the optimization mindset changes people. In an experiment by a team at Shanghai Jiao Tong University, participants who completed a task with AI assistance and were then given a different task without AI showed an 11% drop in intrinsic motivation and a 20% rise in boredom. AI had made the first task so enjoyable that ordinary work became unbearably dull by comparison.

A study at the University of Pennsylvania’s Wharton School deployed an AI deliberately designed to give wrong answers. 80% of participants accepted the AI’s errors as-is. Without a reference point of your own judgment, whatever the AI says becomes the truth. Brooks calls this “cognitive surrender.”

One scene from Brooks’s column captures this shift perfectly. A school principal in the U.S. showed students a film that 200 artists had spent five years making. One student asked, looking baffled, “Why would you do it that way? AI could do it in five minutes.” That question isn’t about speed. It’s the optimization worldview reaching its logical conclusion: the output matters more than the process.

There’s an important counterpoint here. Not every AI user in the Carnegie Mellon study collapsed the same way. Those who asked AI directly for answers (61%) saw both their ability and their motivation plummet, but those who only requested hints or background explanations showed no difference from the group that used no AI at all. The same tool produced completely different outcomes depending on how it was used. Whether you treat AI as an oracle or as a librarian determines whether your ability is preserved or degraded.

🇰🇷 Why Korea Is More Vulnerable

There are structural reasons why this problem could be especially severe in Korea.

According to the Bank of Korea’s 2025 survey on AI usage, 51.8% of Korean workers use AI for work, nearly double the U.S. rate of 26.5%. The share of heavy users—those spending over an hour a day with AI—is 78.6% in Korea, dwarfing the U.S. figure of 31.8%. In a May survey of 1,000 workers by Now&Survey, the share using AI at least once a week rose to 74.3%. Korean office workers are among the most aggressive AI adopters in the world.

What’s interesting is that the group using AI the most is also the most anxious. In IT and development roles, 87.6% actively use AI and report the highest efficiency gains of any occupation, at 90.7%—yet they also report the highest sense of crisis, at 70.1%. 23.2% of respondents named “people who rely on AI without thinking for themselves” as a risky type, nearly identical to the 22.8% who named “people who reject AI unconditionally.” The idea that blind acceptance and blind rejection are equally dangerous has already taken hold on the ground.

The problem is that this awareness is hard to translate into action in Korea’s environment. In Microsoft’s 2026 Work Trend Index, 78% of Korean workers said “if I don’t adapt to AI quickly, I’ll fall behind”—13 percentage points above the global average. Yet only 16% felt that “our company’s AI direction is clear.” Individuals are anxious, and organizations are directionless.

In PwC’s global workforce survey, only 22% of Korean respondents feel “it is safe to try new approaches at work”—less than half the global average of 56%. When there’s no room to experiment and fail, people have no choice but to lean on optimization. And the more you lean on optimization, the faster the muscle of independent thinking atrophies.

There’s one more data point worth noting. According to an analysis by the Korea Labor Institute, jobs for workers aged 15 to 29 fell by 211,000 over the past three years in industries with high AI exposure. 98.6% of that decline occurred in high-AI-exposure sectors. Meanwhile, jobs for people in their fifties actually increased in those same sectors. Researchers call this “seniority-biased technological change,” because AI first displaces the standardized knowledge work that junior staff typically handle. The generation that most needs the chance to build cognitive muscle is the first to lose that chance. The erosion of thinking ability is also an intergenerational equity problem.

Oswald’s Lens

Looking at this research, I was reminded of a pattern I kept running into while building GTM strategies.

When a company adopts a new tool, productivity clearly rises at first. But the moment the tool starts substituting for thinking, the organization’s judgment quietly erodes. Adopt a CRM, and your feel for customers dulls. Install a dashboard, and you stop visiting the field. Lean on A/B testing, and you stop asking why a result happened.

AI is the most powerful version of this pattern. Earlier tools replaced specific functions, but AI can replace the entire thinking process. So the scope of atrophy is incomparably wider.

Brooks’s frame in this column is concise but sharp. “When intelligence is plentiful, volition is valuable.” He’s saying that an era is coming where it’s not IQ but the willingness to dive into difficult work that determines a person’s worth.

As I see it, this isn’t just an individual problem. It’s also a problem for organizations and education systems. Korea’s workplace environment right now is sending an overly strong signal that says “don’t experiment.” The 22% psychological-safety figure is the evidence. In the age of AI, the most dangerous person isn’t someone who can’t use AI—it’s someone who can’t think without it.

Closing

Just 10 minutes of AI-assisted work is enough to measurably reduce independent thinking and persistence. But when people asked for hints instead of answers, this decay didn’t occur. Korea leads the world in AI usage intensity, yet ranks near the bottom in cultures that tolerate failure. That combination is exactly the environment where “cognitive surrender” can spread the fastest.

In the end, the people who choose the effort of thinking for themselves—even with AI available—will be the ones who stand out. If you’ve ever used AI and felt, “I used to do this myself—why don’t I anymore?” tell me in the comments which area that was.

💬 Tell me one work habit of yours that’s changed before and after using AI · 📨 Share this with someone it could help

References & Further Reading

Primary sources

Grace Liu et al., “AI Assistance Reduces Persistence and Hurts Independent Performance,” arXiv:2604.04721, 2026. — The core evidence behind today’s newsletter. Figure 3’s comparison of “hints vs. direct answer requests” is especially useful.
Nataliya Kosmyna et al., “Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task,” arXiv:2506.08872, 2025. — A sprawling 206-page study. Its EEG-based brain connectivity analysis is the core contribution.
Krzysztof Budzyń et al., “Endoscopist deskilling risk after exposure to artificial intelligence in colonoscopy,” The Lancet Gastroenterology & Hepatology, 2025. — The first empirical study showing AI degrades medical experts’ skill.
ActivTrak Productivity Lab, “2026 State of the Workplace,” 2026. — An analysis of 443 million hours, rich with before-and-after AI adoption comparisons.
David Brooks, “The People Who Will Thrive in the AI Age,” The Atlantic, June 28, 2026. — The starting point for today’s newsletter. The frame “when intelligence is plentiful, volition is valuable” is central.

Background

Bank of Korea Employment Research Team, “Survey on AI Usage,” 2025. — Covering 5,512 Korean workers. The original source for the data showing Korea’s AI usage intensity is double that of the U.S.
Now&Survey, “Report on Korean Office Workers in the AI Era,” 2026. — Covering 1,000 workers, with detailed breakdowns of AI usage and perceived crisis by occupation.
Jinsoo Han and Samil Oh, “Age-Based Employment Changes in High-AI-Exposure Industries,” Labor Review, 2025-2026. — Korean data documenting the decline in youth jobs within high-AI-exposure sectors.

The author, Kwangseob Ahn, is a professor of business administration at Sejong University and lead consultant at OBF (Oswarld Boutique Consulting Firm). He teaches statistics and data analysis — business data management and business analytics — while leading GTM and AI strategy consulting in the field, designing the seam between technology and business. He has published academic research on a memory architecture for AI dialogue systems (HEMA) and runs Daily Arxiv, a daily curation of global AI papers. He holds a master’s from Korea University’s Graduate School of Technology Management and a KMBA. He is the author of Homo Brainless: The People Who Outsource Their Thinking.

Footnotes

Brain connectivity: a measure of how actively different regions of the brain exchange signals with one another. Higher connectivity means the brain is actively processing information; lower connectivity suggests the brain is closer to “power-saving mode.” ↩