There is a version of using AI that makes you sharper, faster and more capable, and a version that quietly hollows out the exact skills employers pay for. Most students cannot tell which one they are doing. With midterms here, this is a good week to find out.

This essay is for students, recent graduates and anyone early in a career who grew up with AI tools within reach. It is not an argument against using them. It is an argument for understanding what you trade when AI does the thinking for you, and why that trade costs far more than it appears to.

The people who will be most valuable over the next decade are not the ones who prompt best. They are the ones who can still think when the prompt is gone.

The Short Version

AI Is Remarkable, and the Distinction Still Matters

Apple cofounder Steve Wozniak has said he believes in the artificial part of AI, not the intelligence part. That is not a dismissal of how powerful these tools are. It is a precise observation about what they do.

AI finds patterns in data it has already seen and produces output that is statistically plausible. It is fast, broad and genuinely useful for organizing, drafting, summarizing and iterating on things that already exist.

Here is the distinction that matters for your career. Using a tool to extend your capability is different from using a tool to skip the work of building that capability in the first place. A calculator is a tool. A calculator you pick up before you learn arithmetic is a crutch that keeps you from understanding the math. The same logic holds for writing, analysis and critical thinking. AI can accelerate someone who already has those skills. For someone still developing them, it can quietly stop them from developing at all.

What AI Cannot Do

AI works on what has already been done. It learns from existing text, existing solutions and existing patterns, which makes it powerful for the known and useless for what has never been done before.

Every genuinely new product, business model and creative breakthrough comes from a human mind that can hold ambiguity and build something that does not yet have a training set. That is the space AI cannot enter. The question is whether you are building the kind of mind that can work there, or training yourself to need a prompt before you can think. If you outsource your thinking during the years you are supposed to be building those cognitive muscles, you arrive at your career with a very smooth interface and very little underneath it.

Why the Dementia Parallel Hit Me Personally

I watched my father lose himself to Alzheimer’s. He was a sharp, intelligent man. Watching a person lose the ability to retrieve what they knew and who they were is one of the hardest things I have ever witnessed. So when researchers use the phrase digital dementia, I do not read it as a metaphor. I read it as a warning.

The mechanism is different. Alzheimer’s is a disease. Cognitive offloading is a habit. But the outcome they share is a mind that can no longer independently reach what it used to know. One happens to you. The other, if the research is right, you do to yourself, slowly, by choosing not to think whenever a tool will think for you. I do not want that for myself, and I do not want it for my students.

What the Research Says About Offloading

German neuroscientist Manfred Spitzer popularized the term digital dementia in his 2012 book Digitale Demenz, arguing that cognitive skills, like muscles, weaken when they go unused. A 2024 review of the research on digital dementia and cognitive impact in the NIH archive reaches a cautious version of the same conclusion: heavy reliance on digital tools is linked to weaker attention and memory.

The newest evidence is about AI specifically. In a study from the MIT Media Lab, 54 participants wrote essays in three conditions: with their own brain only, with a search engine, and with an AI chatbot. EEG readings showed the strongest brain connectivity in the brain only group, a middle level for search, and the weakest for the chatbot. The chatbot users also reported the lowest sense of ownership over their essays and struggled to quote their own writing.

A separate survey of 666 people by researcher Michael Gerlich, published in the journal Societies in 2025, found a strong negative correlation between frequent AI use and critical thinking scores. The heaviest users were the youngest, ages 17 to 25.

Be fair to the evidence. The MIT study is a preprint with a small sample, and the Gerlich survey is correlational, so neither proves that AI causes weaker thinking. Treat them as a warning light, not a verdict. The warning matches what employers tell me. A senior engineer at one of South Korea’s largest automotive companies described it plainly: new engineers often cannot explain the first principles behind the code they submit. They can produce output. They cannot defend it, and they cannot diagnose it when it fails. That is not a technology problem. It is a learning problem created by one way of using technology.

What the Job Market Is Saying

The World Economic Forum’s Future of Jobs Report 2025 projects about 170 million jobs created and 92 million displaced by 2030, a net gain of 78 million. It also expects 39 percent of key skills to change over the same period. The new jobs go to people who can work alongside AI, not people who learned to defer to it.

The entry point is where it is tightening first. A Stanford Digital Economy Lab study of payroll data, Canaries in the Coal Mine, found a 13 percent relative decline in employment for workers aged 22 to 25 in the most AI exposed occupations. Experienced workers in the same occupations stayed stable or kept growing. In other words, the tasks AI absorbs first are the ones that used to teach beginners their trade.

Employers are noticing the same thing from the hiring side. Drexel’s 2026 College Hiring Outlook, based on more than 600 employers, found hiring optimism at its lowest in more than a decade. Employers say they want demonstrable skills, authenticity and applicants who truly own their materials. The window is not closing, but it is moving, and the people competing for those roles are not waiting.

The Companies That Cut, Then Came Back

The idea that AI simply replaces people had a good run. Then reality showed up. Klarna became the most cited example. In 2024 it said its AI assistant did the work of 700 customer service agents. By May 2025 its CEO told Bloomberg the company was hiring human agents again, admitting the AI was cheaper but produced lower quality.

Klarna is not alone. A report in People Matters cites Forrester’s 2026 predictions that 55 percent of employers who cut headcount because of AI now regret it, and Gartner’s forecast that half of the companies that cut customer service staff for AI will need to rehire similar roles by 2027.

The lesson is not that AI failed. AI handles volume. It struggles with nuance, escalation and the edge cases that make up the hard part of the job. The people who get hired back are the ones with deep domain knowledge, strong communication and enough understanding of the problem to know when the AI is wrong. Those skills do not arrive automatically. They are built through years of hard work, feedback and failure, and only you can do that part.

What Failing Taught Me

I have taught communication and career development at a Seoul university for 19 years, and I have lived in Seoul for 24. I have also failed, repeatedly, at things I was sure would work. Job applications I was certain I would get. Presentations where I lost the room completely. Projects I poured months into that collapsed. Events I organized that went nowhere near the plan. Courses that did not land the first time. Proposals that were rejected.

Every one of those failures taught me something success would not have. I learned to read a room, what Koreans call 눈치 (nunchi), because I had lost rooms. I learned to manage a project because I had watched them fall apart. I learned to write because I was told my writing was not good enough and had to work out why. The struggle was not the obstacle to the learning. The struggle was the learning.

I am not sharing this to inspire anyone. I am sharing it because I am watching a generation build professional skills with a tool that removes the productive struggle entirely. I worry about what that costs them at 35, standing in front of a problem nobody has seen before, with no AI that knows the answer.

The Only Test That Matters: Can You Defend It?

There is a simple, reliable way to tell whether a student, an employee or a candidate understands what they produced. Ask them to present it, then ask questions. Not trick questions. The natural follow ups any informed person would ask. Why did you structure it this way? What did you consider and reject? What changes if the constraint shifts? Where is the weakest point in your own argument?

Korean job interviews already work this way. Panels chain one question onto the last, a pattern candidates call 꼬리 질문 (kkori jilmun, literally tail questions). The first answer gets you in the room. The third follow up shows whether the thinking was yours. A person who did the thinking can answer. A person who ran the prompt cannot. If you want to practice, my guide to common English interview mistakes shows where candidates usually slip.

For students, your professors and future employers will use exactly this test. A polished deliverable you cannot explain is not an asset. It is a liability the moment someone asks a question. For employers, stop asking for work samples alone. Ask people to walk you through their thinking. The gap between what someone submits and what they can defend tells you nearly everything.

If the Company Can Prompt It, They Do Not Need You

If your main contribution is output AI can produce, you are competing with a tool that works faster, costs less and needs no benefits. The question is not whether you can use AI, because most people can. The question is what you bring when the prompt runs out: your judgment, your domain expertise, your ability to ask the right question before anyone knows what the right question is, your credibility in a room, your relationships, and your capacity to work in ambiguity.

Those things are built through real work, real mistakes and real conversations. If you want to be the person a company keeps when the budget tightens, be the one who adds something to the discussion that was not already there. Be the one who can explain why the AI was wrong. I made a related argument about authenticity and personal brand earlier this year, and it applies here too.

Use AI After You Think, Not Instead of Thinking

None of this means do not use AI. It means use it at the right stage. Write your first draft before you ask AI to improve it. Form your own analysis before you ask AI to summarize. Develop your own solution before you ask AI to refine it. That sequence keeps the cognitive work that builds skill. The reverse sequence removes it.

Use AI to check yourself, challenge your assumptions and surface what you missed. Use it to iterate faster on ideas you already own and to handle the mechanical parts of work so you can spend more time on judgment. The tool amplifies what you bring. If what you bring is shallow, the output is shallow with better formatting.

A Midterm Week Protocol

Here is how to apply this before your next exam or assignment, within your course’s AI policy. First, write your own answer or outline before you open any AI tool, even if it is rough. Second, close the laptop and explain the idea out loud for two minutes, to a friend or to the wall. Third, ask AI to attack your argument, find the gaps and quiz you, and note what it caught that you missed. Fourth, rewrite the weak parts yourself. If you cannot get through the second step, you do not yet know the material, and an exam room will show it. For presentation nerves specifically, presentation anxiety is a preparation gap, not a personality flaw.

Be Better. Not the AI.

I want to be direct. When I write or talk about this, I am not performing concern. I want the people coming through our classrooms and into the workforce to be strong, capable of hard things, and the kind of professionals organizations actually need, not the kind that get cut when a tool improves. We are passing the torch to a generation that will run a future I hope we can trust and not fear. That is a real responsibility.

So here is my honest offer. If you are building a skill, making a career decision or working out how to use AI well without depending on it, reach out. I mean it. You do not need it figured out. You just need to be willing to do the actual work.

Prove it to yourself first, then to everyone else. If you want a coach in your corner, start here.

Try it now, no signup

The Offload Line

Two questions decide what you hand to the machine and what stays with you. Plot your real tasks and read the shape.

Open the tool →

The Briefing, Twice A Month

One idea, one tool, one thing happening in Seoul. No filler, unsubscribe in one click.

Subscribe free →

CareerCommsTwenty five years in business communication, marketing and branding, twenty four of them in Seoul. Written by a university lecturer and consultant; the frameworks here are the ones we teach with. About · For companies