You do not have to do every task manually to remain capable. You also do not have to delegate a task merely because a tool can produce an answer.

The useful question is more specific: Which abilities do I need to judge the result, recover when something fails, and keep doing work I value?

The answer depends on your role. A designer, an administrator, and a researcher need different depth in different skills. An adult returning to study may also choose a skill for confidence, curiosity, or pleasure rather than immediate workplace value.

This article offers a way to choose. It is a personal planning exercise, not a validated assessment of dependence, employability, or intelligence.

Begin with the work you actually do

List five tasks from an ordinary week. Include small tasks rather than broad job titles: explaining a delay, checking an invoice, reading a report, choosing a supplier, or planning a meeting.

For each task, ask four questions:

  1. What do I need to understand to recognize a poor result?
  2. What must I explain to another person?
  3. What would I do if the tool were unavailable or wrong?
  4. Which part do I want to get better at for my own reasons?

You are looking for responsibilities and interests. You are not trying to become equally expert at everything. A skill worth retaining may still be used with a calculator, reference book, accessible technology, or another person's assistance.

Worked example: a weekly project update

Imagine an operations coordinator who sends a weekly update. An AI tool can turn notes into paragraphs, but several important decisions precede the writing.

The coordinator needs to decide what changed, which delay matters, who needs to act, and whether a claim is supported. They also need to explain the situation when someone asks a question in a meeting.

A practical division might look like this:

Part of the task Practice to retain Assistance that may fit
Identify the main issue Write a three-sentence assessment Ask for questions about missing context
Explain the situation Draft the opening in your own words Request feedback on clarity
Report a number Recalculate from the underlying figures Use a spreadsheet for arithmetic
Recommend an action State criteria and tradeoffs Generate alternatives to examine
Format the update Know the required structure Use a saved template

This is one reasonable arrangement, not a universal boundary. If your role requires strong writing, you may choose to draft the entire update yourself regularly. If writing is an access barrier, appropriate assistance may be essential while your practice focuses on selecting and communicating the meaning.

Protect the abilities that let you inspect the work

For many information-based roles, three kinds of practice are useful starting points.

Explain a claim. Put the main point in ordinary language. Identify the evidence and what remains uncertain. If you cannot explain a recommendation, you may be missing information needed to own the decision.

Estimate and calculate. Make a rough prediction before accepting a result. Rebuild a percentage, total, or comparison from its inputs. You do not need to become a mathematician to notice that a tenfold change deserves inspection.

Choose and justify. Name the criteria that matter, compare at least one alternative, and explain the tradeoff. “The tool preferred this option” does not describe why it suits your circumstances.

These are editorial recommendations. They are not a claim that every person needs identical practice or that using assistance inevitably weakens ability.

Treat research as a reason to investigate your habits

A CHI 2025 survey asked knowledge workers about their AI-assisted work. Greater confidence in AI was associated with less reported critical-thinking effort. The study used self-reports, so it cannot establish that a particular person's skills declined or that AI use caused permanent impairment. Read the primary publication.

That finding is a prompt for a useful personal question: “When an answer looks convincing, what do I stop checking?” Your answer may concern citations, arithmetic, tone, missing context, or the decision itself. Choose a practice that addresses the gap you actually observe.

Make the practice small enough to repeat

Pick one skill for the next two weeks. Attach it to real work and define an observable action.

Instead of “improve my thinking,” try: “Before each weekly update, I will spend five minutes writing the problem, evidence, and recommended action without generated text.”

Instead of “stop relying on AI,” try: “Twice this week, I will solve a small spreadsheet problem before consulting documentation, then explain the correction.”

Keep one early attempt. After a few sessions, try a related task before viewing your notes. Compare what you can explain, where you pause, and what help you need. These observations can guide practice; they are not scientific proof of improvement.

Count the full cost of assistance too. Generating, checking, and repairing a draft can take longer than starting from your own outline. On another task, assistance may be worth the effort. Record the difference instead of assuming a single answer applies everywhere.

You can also choose something because it matters to you

Writing a personal letter, reading a novel slowly, drawing, or solving a puzzle may deserve time even if automation could imitate the result. Enjoying the activity is a sufficient reason to keep doing it.

You do not owe a productivity argument for every preference. Choosing not to use AI for a particular task is compatible with using other technology and asking for human help.

For today's exercise, choose one recurring task and complete this sentence:

I will keep practicing ______ because I want to be able to ______. This week I will practice by ______. I will use help when ______, and I will check ______ myself.

Read it back. Is the practice concrete, realistic, and connected to something you value? If so, you have a starting point. Review it after two weeks and decide what deserves another round.