How to Use AI Without Losing Your Own Skill

The risk nobody warns you about

Most advice about AI at work worries about the wrong things: accuracy, privacy, whether the tone sounds robotic. Those are real and they are also visible, which means someone will catch them.

The risk in this lesson is different, because nothing about it looks like a problem while it is happening. It is this: if the tool does all of your thinking, your own capability quietly erodes, and the erosion is invisible until the day you need it.

The mechanism is not mysterious. Skill is maintained by use. The uncomfortable part of any task, the part where you sit with a blank page and force yourself to work out what you actually think, is not an inefficiency in the process. It is the process. It is the bit that built the skill in the first place, and it is precisely the bit that is now optional.

In plain English

Deskilling:: The gradual loss of an ability through not using it. It happens quietly, and you usually notice only when you need the skill back.
Fluency illusion:: Feeling that you understand something because you watched it being done smoothly, rather than because you can do it.
Effortful retrieval:: Pulling something out of your own head before looking it up. Uncomfortable, and one of the strongest ways to learn.
Load-bearing skill:: A skill your job actually rests on, as opposed to one that is merely on your CV.

Why this matters more than it sounds

Two reasons, and the second one is the serious one.

The first is obvious. If your capability erodes, you become less valuable in the ordinary way, and you notice it in interviews and in hard weeks.

The second is circular and much worse. Everything in the last lesson said the durable position is being the person who can check the output. But you can only check what you could, in principle, have produced yourself. Hand over the producing entirely and you lose the ability to check, which was the whole plan. You become a person who forwards output, and forwarding is not a job.

The dangerous version of this is comfortable. Your work goes out faster, nobody complains, and the quality looks fine because the tool is good. That is exactly the condition under which the erosion goes unnoticed for a year.

Four habits that keep the skill

None of these require you to use AI less. They require you to use it in a different order.

1. Attempt first, then compare

This is the single highest-value habit in this course. Before asking for anything, spend a few minutes producing your own version. It can be rough. It can be five bullet points and a sentence about what you are unsure of. Then ask the AI and compare the two.

The comparison is where the learning is. You will see what it caught that you missed, and, more often than people expect, what you had that it did not. The few minutes of your own effort is what converts the AI output from a substitute into a lesson.

Prompt you can copy

I attempted this task myself first. Below is my attempt, then the brief.
Do not rewrite my version. Instead:

What did I miss that a strong version would include?
What did I get right that you would not have thought to include?
What is the single habit of thinking that separates my attempt from a strong one?
MY ATTEMPT:
[paste]
THE BRIEF:
[paste]

2. Use it to critique rather than to produce

Asking for a critique keeps you as the author. Asking for a draft makes you an editor of someone else's thinking, and editing exercises a much narrower set of muscles than writing does.

❌ Weak prompt

Prompt

Write the client proposal for the office refit project.

Output

A polished four-page proposal with sections, pricing structure and next steps, delivered in seconds.

Fast, and you learned nothing. You now have a document you did not think through and cannot fully defend in the meeting.

✅ Good prompt

Prompt

Here is my draft client proposal. Attack it as the client's finance lead would. What is unclear, what looks like padding, and what would make you say no? Do not rewrite anything.

Output

The pricing in section 2 hides the contingency, so it reads as if you are being vague on cost. The timeline assumes access you have not confirmed.

You keep authorship, you get the pressure-testing, and next time you write it better yourself.

3. Keep doing the hard part by hand, sometimes

Pick the one or two things your job genuinely rests on, your load-bearing skills, and do them unassisted on a regular basis. Not everything, and not always. Once a week is enough to keep the machinery warm.

For most people this means the thinking-heavy task rather than the tedious one. Let AI have the formatting, the reformatting, the first-pass summary and the boilerplate. Keep the analysis, the argument and the judgement.

Prompt you can copy

Here is a list of the recurring tasks in my job.
Split them into two lists: tasks where handing the work to AI costs me nothing important, and tasks where doing it myself is what maintains a skill my job depends on. Explain each placement in one line.
Be strict. If a task builds judgement or domain understanding, it belongs in the second list even if AI could do it well.
MY TASKS:
[paste]

4. Test yourself occasionally

Once a month, try to explain something in your field from memory, without looking anything up, then check yourself. Effortful retrieval is uncomfortable and it is the fastest way to find out that something you believed you knew has gone soft.

Prompt you can copy

I am going to explain a concept from my field from memory, with nothing open in front of me.
Afterwards, tell me: what I got wrong, what I stated too vaguely to be useful, and what an expert would have included that I left out. Be direct. Do not be encouraging.
THE CONCEPT: [name it]
MY EXPLANATION FROM MEMORY:
[paste]

There is a useful test for any AI habit. Ask yourself whether, after a year of doing it this way, you would be better at your job or merely faster at producing output. Both are legitimate answers, but you should know which one you have chosen.

Checkpoint

Skill is maintained by use, and the uncomfortable part of a task is the part that builds it. Attempt first then compare, use AI to critique rather than produce, keep your load-bearing skills unassisted, and test yourself from memory.

The trade-off, stated honestly

All of this is slower. Attempting first costs you fifteen minutes. Doing the hard thing by hand once a week costs you an afternoon. Anyone telling you these habits are free is selling something.

What you buy with that time is the ability to still be good at your job in a few years, and the ability to check work you are putting your name to. Whether that is worth fifteen minutes is a judgement call, and it is yours. Make it deliberately rather than by drifting into the fast option every single time.

📝 Quiz

Question 1 of 4

What is deskilling, in the sense used here?