The question I get asked most often is some version of: “Will AI take my job?”

I build AI systems for a living. And I’ve started answering it differently than I used to.

The problem with the question isn’t the answer — it’s the assumption built into it. It assumes your job is one thing. It isn’t. Your job is a bundle of tasks, and AI isn’t coming for all of them. It’s coming for some of them. The honest question is which ones.

Break your job apart

Take a blank sheet of paper. Write your job title at the top. Then list every task that title actually involves on a normal week.

If you’re a lawyer, that list includes drafting contracts, reviewing case law, preparing discovery, writing memos, taking client calls, negotiating terms, reading the room in a deposition, giving bad news to a client, mentoring a junior associate, managing a partner’s expectations.

If you’re an accountant, it’s reconciling books, preparing returns, translating tax rules into plain English for a client, noticing something odd in a ledger, making the client feel less anxious about an audit, telling them when they should see a lawyer instead.

If you’re a marketing director, it’s writing briefs, reviewing creative, approving budgets, running quarterly reviews, sitting with a CEO who wants pipeline yesterday, deciding what not to do this quarter, hiring, firing, saying no.

Most job-title lists look like that. A few tasks are the ones people think the job is. The rest are the ones the job actually is.

Now sort them

Next to each task, write one of three things:

  • Routine — anyone who’s done this role for a year can do it, with enough time.
  • Judgment — doing it well requires you to have met this client, read this room, lived through this quarter.
  • Trust — someone is counting on you, specifically, to own the outcome.

AI is coming fast for the first list. It’s already here. If a task is routine and language-shaped — summarise, draft, search, compare, format — a tool exists today that does it in a tenth of the time. Three years from now that tool will be free and built into the software you already use.

The second list is where it gets interesting. AI can help, but it can’t own the call. The judgment comes from your context — things you know that haven’t been written down.

The third list AI can’t touch at all. Not because it lacks capability, but because trust isn’t a capability. It’s a relationship.

The real question

The question isn’t “will AI take my job.” It’s:

What fraction of my job is on the routine list — and what am I going to do with the hours AI frees up there?

If it’s 70% routine and you don’t move, in three years you’ll be doing 30% of a job that still wants to pay a full salary. That’s a hard position.

If it’s 70% routine and you push every one of those hours into the judgment and trust lists, you’ll be doing the best 30% of your current job with double the time for it. That’s a great position.

The people who thrive through this transition aren’t the ones who resist the routine getting automated. They’re the ones who noticed which parts of their work they actually wanted to do more of, and made space for it.

One thing to do this week

Do the sort I described. Not in your head — on a page. Look at your three lists.

You’ll see what you want to protect. You’ll see what you can probably hand off. And you’ll see the shape of the job you’d rather have three years from now.

Then the only question left is whether you’re spending this week practicing for that job, or defending the one AI is already taking half of.