What Should I Learn in the AI Era? Start With Two Lists

Everyone says learn AI; nobody says what. Describe one weekly task to AI and split it in two lists: what to hand over, and the skills worth sharpening.

On this page

Everyone says "learn AI". Nobody tells you what. The usual answer is a list: prompt engineering, Python, data literacy, a certificate or two. It is the same list for everyone, which is the problem: it isn't about your work. There's a shorter way to find out what to learn, and it takes ten minutes. Describe a task you repeat every week to AI and ask which parts it can do and which need your judgment. You get two lists. Hand the first one over. The second one is your skill list.

The step: one task, two lists

Pick a task you do every week: a report, a reconciliation, a review, a planning round. Describe it to your AI assistant the way you'd explain it to a new colleague, including the awkward parts. Then ask:

TEXT
Which parts of this can you do, and which parts need my judgment?

Read both lists and correct them. AI will put a few things on the wrong side; moving them is the first useful thing you'll learn.

What to do with the first list

The first list is the work AI can take. The skill you need here is not "AI" in general. It is two narrower things:

  • Describing the task well. The context, the constraints, what a good result looks like. Most bad output is a vague request.
  • Checking the result. Knowing where it tends to be wrong in your kind of work. If you've done the two-line test, you already know where to look.

Hand one item over this week. Not all of them; one, checked properly, teaches more than ten done carelessly.

Why the second list is the one that matters

It is made of your work, not of a trend. The skills on it are the ones your role depends on today, which is why they are worth sharpening: deciding between options, explaining a trade-off, knowing which exception matters, getting people to agree.

It tends to be the senior half of the job. When the routine part moves to AI, what's left is closer to judgment and communication. That's also the part people notice when they decide who leads next, which is why the same habits show up in an update your manager acts on and in showing you're ready to manage.

It shows you where a product might be. A step you do by hand every week, that AI can't do and nobody else in your company does either, is sometimes a need nobody wrote down. The same idea, applied to a spreadsheet, is in where your next product idea is hiding.

The split between the lists isn't fixed. Some tasks are best done by AI on its own, some with AI helping you, a few by AI deciding inside limits you set; automation, augmentation and autonomy is that distinction worked out for enterprise processes. For a whole workflow rather than one task, the free Workflow X-Ray maps the hidden decisions, the manual work and what could be automated.

Where it goes wrong

  • Describing the ideal task instead of the real one. If you leave out the exceptions, AI will claim the whole task. Put the bad day in.
  • Believing the first split. AI is generous about what it can do. Try one item from the first list before you hand the rest over.
  • Learning everything on the trend list anyway. The point of the second list is permission to ignore most of the rest.
  • Pasting what shouldn't leave the building. Describe the task, not the data. Use the assistant your company approves.

Don't learn everything. Learn the second list.

Share

Frequently asked questions

(3)

What skills should I learn for the AI era?

Two kinds, and your own work tells you which. Describe a task you repeat every week to an AI assistant and ask which parts it can do and which need your judgment. For the first list, the skill is handing work over well: describing the task clearly and checking the output. The second list is the skills worth getting better at, because they are the parts that stay yours.

Do I need to learn to code to stay relevant?

Not as a rule. If the task you hand over to AI needs a small script or a spreadsheet formula, you'll pick up as much as the task needs, with AI writing most of it. What matters more is being able to say exactly what you want and to tell when the result is wrong, and that comes from knowing the work, not from knowing a language.

Is prompt writing a real skill or a fad?

The tricks will date. The underlying skill won't: explaining a task clearly enough that someone else can do it, with the context, the constraints and what a good result looks like. That is what good managers have always done with people, and it is most of what makes a prompt work.

The daily step goes out on LinkedIn and X.