There Are No AI-Resistant Skills

A coffee shop in Berkeley, 2pm on a Tuesday. Wooden tables dented by a decade of laptop work. Stefanie is 23, just finished her internship at McKinsey's San Francisco office, full-time offer in hand for next month. She opens a document. Seventy-five pages. Market analysis for a European telecom: competitor landscape, pricing dynamics, regulatory environment, customer segments, growth projections. The kind of work that took senior consultants weeks, that McKinsey billed at EUR 50,000, that became the foundation for a EUR 200 million capital decision. She wrote it in fifteen hours. I asked her to hand the same problem statement to Claude. Just the brief. No guidance. We waited. Eleven minutes. Identical structure, identical competitor breakdown, identical regulatory reading, identical conclusions. The document that was going to pay her EUR 500,000 over a career cost less than one cent. She stared at the screen for a long time and said nothing. "It's not worse," I told her. "It's the same. That's the part you have to absorb."

This did not start with AI. It started in a London studio in 1979. To release a song in 1980 you needed a band, months of rehearsal, studio time at roughly GBP 800 a day at Abbey Road, an engineer, a producer, mixing, mastering, a label willing to front GBP 50,000 to GBP 250,000, a distribution deal, and a vinyl plant. Eighteen months from written to heard, and a half-dozen gatekeepers who could end your career with one phone call. The barrier was brutal. The barrier was also the price. Scarcity in, margin out. In 2024, Spotify took in about 100,000 new tracks a day. That is 36.5 million songs a year. Every recording made between 1900 and 1990, ninety years of it, comes to roughly 4 million. Spotify passed that in six weeks. Around 40% of the tracks on the platform have never been streamed once. A sixteen-year-old with GarageBand and a USD 30 DistroKid account is on every platform by morning. No engineer, no producer, no budget. Same collapse in publishing. About 50,000 books were published in the United States in 1980. Amazon's self-publishing side carried roughly 4 million by 2023. Most sell fewer than ten copies. The probability that this is a temporary oversupply is not zero. Most people are pricing it at zero. Supply did not rise because quality rose. Supply rose because the gate stopped being expensive.

The guidance counsellor handed you a laminated pamphlet. Pursue work that requires human creativity and human empathy. Fundamentally human. Fundamentally about connection. That pamphlet is a list of the highest-margin jobs a machine can now do. AI does not struggle with creative work. Creative work happens on a screen, inside language and data, which is exactly where these systems are strongest. AI struggles with a pipe under a kitchen sink and a truck in the rain. Embodied work is safe. Screen work is not. Legal analysis: AI has passed bar exams in multiple jurisdictions at the 90th percentile, not the 50th. Medical diagnosis: systems trained on twenty million scans now catch tumours experienced radiologists miss, in clinical trials across fourteen countries. Financial analysis: the EUR 50,000 report is an eleven-minute task. Consulting: a twelve-week, EUR 300,000 engagement built from one hundred hours of senior analysis and two hundred of junior work, delivered from a prompt. None of that is a forecast. It is the current state of a market that prices on scarcity, and the scarcity just moved.

People bring me the same four plans. "I'll become a prompt engineer." That was a career in 2023. Modern models rewrite vague instructions into better prompts than the person asking. Half-life: eighteen months. That is a transition skill, not a job. "I'll lean on emotional intelligence." A system trained on billions of recorded human interactions knows when to push and when to soften. You will not out-empathise it on average. "I'll move into management." Management is coordination, prioritisation, allocation, review. That is exactly what an orchestrated agent stack does, without sleeping, without favourites, without forgetting what someone said three months ago. "I'll go somewhere regulation protects me." Regulation is written by the same firms that hire the consultants. The probability that you are on the wrong side of this is not zero, and if you are doing structured knowledge work the way the system trained you, it is closer to sixty percent.

The skills you can describe are the skills that get replaced. The job description was the notice.

A high-school senior, seventeen years old, learned to code at eight. She arrived with a portfolio of bad projects: buggy, badly structured, half of it did not run. I gave her the challenge I give everyone. Build a compliance tool, full-stack, three weeks, you have every tool available, I am not here to guide you. Week one, frustrated. Week two, she stopped asking permission and started asking better questions. Week three she shipped a working deployment, better documented than what we started with. Not because she codes well. Thousands of people code. Because she could orchestrate tools, decide what to build, and iterate on what came back. She moved from coder to creator in three weeks, at seventeen. That is the only line that matters now. Creator or consumer. I have built companies across twelve countries, taken a company through a Frankfurt IPO, deployed 210 energy systems, restructured a EUR 75 million industrial group. I have also signed a EUR 20 million personal guarantee and watched a company fold on Christmas Day 2008. The guarantee taught me more than the IPO did, and the lesson was not about credentials. There were no refunds on the credential. There is no refund on yours either.

One category survives, and it is not a skill in the way you were taught to think about skills. Trust. Not competence trust. Relationship trust. The irrational, accumulated, micro-signal kind built from shared meals and shared failures over years. The reason one phone call from Keith Ferrazzi opens a door that EUR 500,000 of consulting fees cannot. The reason a bartender in Florence could tell me at 23 that there is always room at the top and have me rearrange my life around it. Here is the test. The moment you can state clearly what you do, you have written the job description for the machine. The work that survives is the part you cannot fully articulate: the read of a room, the judgment call with no derivation. A hundred thousand songs land on Spotify today. Almost none will be heard twice. But the one your friend sends you with the words "I made this thinking of you" gets kept for life. Not because it is better. Because it was made by a person who knew you. The probability that your current path prepares you for the next ten years is maybe twenty percent. The probability that you can still move is much higher, but it drops every quarter you wait. So build the portfolio instead of the pedigree. Work on the thing you can point to. Be the person someone trusts when it matters, because that is the one asset a machine cannot buy, scrape, or train. I learned the urgency the expensive way. I left my daughter Luisa to chase a career that felt necessary, and I won a game that was already ending. The window is open for you now. It does not stay open.