The truths nobody told you

A job title is a label on a bundle. A junior marketer writes briefs, cleans spreadsheets, sits in meetings, calms a nervous client and decides what to cut. AI is good at some of that bundle and weak at the rest. “AI will take your job” skips the bundle entirely.

So the realistic threat is rarely “zero jobs.” It’s a job where the parts you were paid for shrink, and the parts nobody automated become the whole job. That shift can help you or hurt you. It depends on whether you see it coming.

In 1964, with automation fears running high, President Lyndon Johnson set up a national commission on technology and jobs. Its 1966 report concluded that automation was not destroying employment: machines were ending particular jobs, not the need for work. Sixty years on, the record still backs it.

ATMs tell the same story. Economist James Bessen found that from 1988 to 2004, the average urban bank branch went from 20 tellers to 13. Cheaper branches meant more branches, up 43% in cities, and teller jobs did not shrink as ATMs rolled out. The job changed instead: less cash handling, more talking to customers.

Now the honest part. “Work survives” describes economies, not individuals. The US Bureau of Labor Statistics projects teller jobs to fall 13% from 2025 to 2035 as more customers bank online, and real people sit in those seats. History says new work will appear. It doesn’t say the work will come to you.

Read that again. The experienced people are mostly fine. The door they walked through is narrower. Entry-level work was always a bundle of simpler tasks (summaries, first drafts, basic code, routine tickets) that employers paid beginners to do while they learned. AI now does many of those tasks, so the training ground shrinks.

This is not a verdict on you. It’s a change of route. If the ladder lost its first rung, build a step of your own: work people can see, on problems people actually have. Parts 6 and 7 show you how.

Picture AI’s ability not as a smooth wall but as a jagged coastline. One task falls inside the line, and the result is fast and impressive. A neighboring task that looks just as hard falls outside, and the AI fails while sounding exactly as sure of itself.

The consultants who got it wrong weren’t careless. They trusted a tool that had been brilliant five minutes earlier. The danger isn’t the obvious error. It’s the fluent one. And the edge keeps moving as models improve, so last year’s map won’t do.

A writing experiment published in Science found the same shape. Shakked Noy and Whitney Zhang recruited 453 college-educated professionals and gave half of them ChatGPT. Time spent on writing tasks fell 40%, quality rose 18%, and the gap between stronger and weaker writers narrowed.

Good news if you’re new: you can produce decent work sooner than any generation before you. The bad news is the same sentence: so can everyone else. When the floor rises for everyone, “decent” stops being a reason to hire you.

Economics has a simple rule: when something gets cheaper, its complements get more valuable. When coffee gets cheaper, people buy more cream. AI makes drafts, options and predictions almost free. What completes them? Someone who can say which one is right, and why.

Make it concrete. Anyone can now generate fifty logo drafts in minutes. The client still needs one. The person who can pick it, defend it, and explain why the other forty-nine fail is doing the scarce part of the job. That’s the part that gets paid.

What leaders say in surveys isn’t always what they do. Still, the direction is clear: the bar moved. Picture two junior analysts. One spends the day cleaning data by hand. The other uses AI to clean it in an hour, checks it, and spends the afternoon on what the numbers mean. Guess who gets the next project.

Notice what that answer does. It names a real task, a result and a check. That’s the new bar: not “I use ChatGPT,” but proof that you use it with judgment.

Psychologist Philip Tetlock spent about twenty years scoring tens of thousands of expert forecasts on politics and economics. On average, the experts did little better than chance, and no better than simple rules like “assume current trends continue.” The better forecasters, whom he called foxes, drew on many ideas and admitted uncertainty.

Hinton is brilliant. He was still wrong, partly because the job is far more than reading images: one 2012 study found radiologists spent only 36% of their time interpreting them. So swap the question. Stop asking “What will happen?” and start asking “What holds up if I’m wrong?”

• Skills that compound: writing, numbers, judgment, learning fast.
• People who know your work and would vouch for it.
• Savings that buy you time to adjust.
• Proof of work that anyone can check.