When to leave AI closed

The fastest reflex is not always the best one. Open the question, bring in the model, read the answer, adjust, move on.

That feels productive. It often is productive. But there is a cost underneath the reflex: you start less often by yourself.

I notice it in myself. Writing SQL. Thinking through an architecture problem. Building an analysis without asking for a first draft. The skill is still there, but switching it on takes more effort than it used to.

That is the risk.

The research is uncomfortable

Gerlich found in 2025, with 666 participants, a strong negative correlation between frequent AI use and critical thinking: r = −0.68. The link runs through cognitive offloading. The more often you place the thinking outside yourself, the less you train the skill.

A correlation is not a verdict. One study is not a law. But the mechanism is recognisable enough to take seriously.

Critical thinking is not a fixed trait. It is a skill. You maintain it through friction: forming a first position yourself, ordering arguments yourself, connecting loose information yourself.

If you skip that friction by default, the skill becomes less available when you need it.

The order matters

The problem is not that you use AI. The problem is that you use the model too early.

Ask first. Read. Adjust. That looks like thinking, but you have not taken your own position. You have edited a given one.

The better order is simple:

1. Write your own answer first.
2. Put your assumptions next to it.
3. Bring the model in after that to attack, improve, or add.

Then AI stays a sparring partner. Not the starting point of your reasoning.

Create no-AI zones

No-AI zones are not purity. They are maintenance.

Start some pieces yourself. Read the paper instead of the summary once in a while. Write a query without autocomplete. Debug code without a copilot. Not because it is faster, but because you keep the muscle under tension.

The same applies to teams. If you only measure whether people get faster, you miss an important question: do they keep thinking better too?

Working well with AI therefore does not mean using AI as often as possible. It means knowing when the model should be open and when it should stay closed.