AI won't steal your job. It takes away the first rung.

On AI and work you hear two opposite stories. One: within a few years agents will do almost all office work and unemployment will explode. The other: it’s always been like this, every technology destroys some jobs and creates others, relax. After reading and listening a lot (and working every day in a firm where AI has genuinely arrived) I think both miss the point.

Tasks, not jobs

The most solid thesis, argued among others by Andrew Ng, is that AI automates tasks, not whole jobs. A job is a bundle of tasks: some repetitive, others about judgement, relationships, responsibility. AI takes the former and leaves the rest.

In my experience that’s exactly right. In a tax firm AI reads invoices and proposes the entries, sorts email, finds the right circular, drafts the first version of an opinion. It doesn’t decide whether a cost is deductible, doesn’t explain to the client why they’re paying more this year, doesn’t sign the tax return. The accountant doesn’t disappear. How they spend the day changes.

The real problem: the first rung

Here’s the part that worries me, and it concerns my generation. The tasks AI does best are exactly the ones young people used to learn the trade.

The trainee logged invoices, and that’s how they understood a balance sheet. The junior lawyer did case research, and that’s how they learned to reason about cases. The junior analyst built the spreadsheets, and that’s how they understood the client’s business. Boring tasks, but they were the gym.

Erik Brynjolfsson at Stanford studies exactly this: hiring data show AI’s impact concentrating on younger workers and entry-level roles, while experienced workers benefit. It makes sense: experience is precisely what you need to check AI’s work.

So the risk isn’t so much mass unemployment tomorrow. It’s a world where companies stop training juniors because “AI does it anyway”, and in ten years find themselves without the experienced people needed to supervise it.

What I’d do as a company

Don’t stop hiring young people. Change what they do. Instead of logging 300 invoices, the junior supervises the automation that logs them, checks the exceptions, explains why one is wrong. They learn more, faster.
Make learning explicit. “Learning the ropes” used to teach by itself. Now it has to be designed: mentoring, work review, discussed cases.
Use AI as a tutor, not a shortcut. The same technology that removes basic tasks can explain, ask questions, correct. But only if you use it that way.

What I’d do as a young person (and I am one)

Become the person who checks. Checking an AI result requires understanding the subject better than someone who merely produces it. It’s the skill that will be worth most.
Invest in judgement and relationships. Deciding, explaining, persuading, taking responsibility: these are the tasks that stay human.
Don’t delegate the effort of understanding. Cal Newport and many learning researchers say it clearly: skills you don’t exercise atrophy. Use AI to remove mechanical work, not the effort of thinking.
Get good with the tools, but not only. Using AI is necessary and not sufficient: in a few years everyone will. The difference will be what you know beyond the tool.

And new jobs?

Daron Acemoglu, Nobel laureate in economics, stresses a point I share: the direction of technology isn’t destiny. AI can be used to replace people or to make them more capable: a technician who diagnoses faults with AI that used to need an engineer, a nurse who handles more patients more safely. The second path creates more value and more work, but it has to be chosen, including by public policy.

In my small way it’s the choice I make in my work: AI drafts, I check. The time saved doesn’t go into “doing more with fewer people” but into doing better: more checks on the numbers, more time with clients, faster answers.

In one line

AI won’t steal your job: it removes the first rung of the ladder. It’s up to those who train, hire and study to build a new one.

Sources

Informational content, current as of the date shown. Prices, models and rules change often: always check official sources. Written with the support of AI tools and reviewed by the author.