Answers

Which parts of my job can AI actually do?

AI reliably does the parts of your job that can be fully specified before the work starts. If you could hand a task to a capable stranger with nothing but a written brief and expect a usable result, that task is exposed. The parts that resist automation are the ones where the brief cannot be written in advance, because the judgment is in deciding what the brief should say.

The question to ask is not about your job. It is about your week.

Most coverage of this asks whether a job title is safe. That framing cannot help you, because nobody is replacing a title. Work gets unbundled task by task, and two people with identical titles can hold completely different exposure depending on how they actually spend Tuesday.

So the useful unit is the task, not the role. A week of work is a stack of tasks, some of which are specifiable in advance and some of which are not. Your exposure is the ratio between them, and that ratio is something you can measure yourself.

The test: can the brief be written before the work?

For each recurring task, ask whether you could write a complete brief for it in advance, hand that brief to a competent stranger with no history at your company, and get back something you would ship.

If yes, the task is specifiable, and specifiable work is where automation is strongest. Drafting a first version, summarizing a long document, converting one format into another, producing a standard analysis from clean data, writing routine code against a clear specification.

If no, ask why not. The answer is almost always that something unwritten has to be supplied: which exception matters this time, which stakeholder will actually block it, which of two defensible options fits a history the brief does not contain. That supplied judgment is the part that does not transfer.

What consistently survives the test

Across most professional work, four kinds of task keep failing the brief test, and they are worth naming precisely because they tend to be invisible on a résumé.

  1. 01Deciding which problem is the real one. Being handed a stated problem and recognizing it is a symptom of a different problem is a judgment made before any brief exists.
  2. 02Knowing which exception matters. Rules cover the common case. Knowing which of this quarter's twelve edge cases will actually cost money is pattern recognition built from having been wrong before.
  3. 03Carrying trust that is specific to you. A relationship where someone takes your read on a call without re-litigating it is not a transferable asset, and it is frequently the reason a decision moves at all.
  4. 04Taste under ambiguity. Choosing between options that are all defensible, where the criteria were never written down, and being accountable for the choice afterwards.

Why your résumé hides exactly the part that matters

A résumé is a list of outputs. Shipped this, managed that, grew this number. Outputs are the most specifiable thing about your work, which means a document built entirely from outputs is a document built entirely from the automatable half.

The judgment sits underneath, in the decisions that produced those outputs, and standard résumé advice actively strips it out in favor of tighter bullets and stronger verbs. That is why people who are genuinely hard to replace often cannot explain why, and why they interview badly for the thing they are actually good at.

Run the audit on your own calendar

This takes about forty minutes and needs nothing but your last two weeks of calendar and sent mail.

  1. 01List every recurring task from the last two weeks. Aim for fifteen to twenty. Be specific: not 'meetings' but 'the Thursday pipeline review where we decide what to cut'.
  2. 02For each one, write the brief you would hand a capable stranger. Time-box it to two minutes each.
  3. 03Mark the ones where you could not finish the brief without writing 'it depends' or 'you would have to know that...'. Those are your non-specifiable tasks.
  4. 04For each marked task, write one sentence naming the specific thing you supplied that the brief could not. This sentence is the useful artifact. It is usually the first clear language anyone has put to it.
  5. 05Count the marked tasks as a share of your week. That share is your honest answer, and it is far more informative than any published list of at-risk occupations.

Where this stops being enough

This audit tells you where your leverage sits. It does not tell you what that leverage is worth in the open market, which is a separate question with a separate answer, and the two are often badly mismatched. A skill can be genuinely rare and still be poorly paid in the room you happen to be standing in.