Practice five habits that keep you thinking while using AI

An impressive answer is useful only if you understand what makes it right. Thiago’s video reflects on a familiar temptation: accepting an AI-generated report because it looks polished. His five habits turn AI use into a chance to practice reasoning and check your understanding.

This guide adapts the practical exercises from the video. It does not claim that AI has been proven to reduce everyone’s intelligence, and you do not need to accept that claim to use the exercises.

What you’ll learn

• Form an initial view before asking for an answer.
• Request disagreement that helps you examine assumptions.
• Verify the facts that support a conclusion.
• Test your understanding without the chat open.

1. Think before you prompt

Take about thirty seconds to write your own first answer. It can be rough. Identify the question, your tentative conclusion, and the reason you currently believe it.

Then ask AI to compare approaches:

Here is my initial answer and reasoning: [text]. Compare it with another plausible approach. Identify where the evidence supports my reasoning, where an assumption is missing, and what I should check next. Do not simply agree with me.

Keep the original note. After reviewing the response, record which part of your view changed and why. That makes the exercise about reasoning rather than just collecting a better-looking answer.

2. Ask for a serious challenge

Use a real decision or argument you are working on. Ask for the strongest counterargument, the weakest assumption, and the evidence that would change the conclusion.

A useful challenge should be specific enough to investigate. “Consider the risks” is vague; “the estimate assumes every team member is available next week” points to a check you can perform. Push back if the assistant invents objections or treats every view as equally supported.

Write your response to the strongest objection yourself. The goal is to improve the argument or revise it, not to make the assistant declare you the winner.

3. Verify the evidence that matters

Choose the claims on which the conclusion depends. Open the original source behind a statistic. Compare a summary with the relevant passage. Recalculate important totals using the actual inputs and an appropriate calculation tool.

Check that a source exists, says what the answer claims, and applies to the situation. A real reference can still be outdated or irrelevant. If you cannot verify a central point, qualify or remove it rather than leaving confident wording in place.

For research across several reports, the NotebookLM briefing guide provides a practical source-comparison exercise. Citations help you find evidence; reading them is still part of the work.

4. Close the chat and explain the idea

After learning something, put the response away. Explain the concept in your own words, give an example, and state a limitation. If you cannot do that, identify the specific part you do not yet understand.

The video suggests recreating an Excel formula or explaining a project-management concept without reading the answer. Use a similar test for your task. Change one input and predict what should happen before checking the result.

Return to the source or ask a focused question about the gap. Repeat the explanation afterward. Memorizing a sentence is less useful than understanding how to apply it in a different example.

5. Ask AI to coach your attempt

Request a problem at an appropriate level and withhold the solution until you have tried it:

Act as my learning coach for [skill]. Give me one practical problem using [context]. Let me attempt it before showing a solution. Then evaluate my reasoning, point out the first important error, and give me a hint so I can try again. Explain the solution after my second attempt.

Choose a skill whose result you can check, such as a spreadsheet calculation, a short explanation, or a structured meeting summary. AI feedback can also be mistaken, so compare technical answers with dependable references or a known result.

Make the habit small enough to repeat

For your next work task, save three things: your first attempt, one correction made after checking evidence, and a short explanation written without AI. Those artifacts show what you learned more clearly than the length of the conversation.

You can still use AI to save time. The aim is to retain responsibility for the conclusion and build the ability to recognize when an answer is incomplete or wrong.