Use AI Without Losing Your Edge

Part three of a series. The first piece made the argument, the second showed the evidence, and this one is the fix.

The first piece argued that AI can turn a tool-user into a tool. The second showed that the decline in reading and attention is real and started before AI arrived. This is the part where we do something about it, and the good news is that the fixes are small and specific. You do not need a digital detox. You need a few habits that keep your brain in the work instead of handing it away.

Can you use AI without it dulling your thinking?

Yes, and the whole game comes down to one distinction: scaffold versus substitute. A scaffold helps you do the thinking. A substitute does it instead of you. The same tool can be either, depending on when you reach for it. The reason the order matters is one of the most settled findings in how people learn. The psychologist Robert Bjork calls them desirable difficulties: the conditions that make learning feel slow and effortful are the ones that make it stick. Anything that makes the work feel easy, including a model that does it for you, tends to produce the weakest memory of it later. Effort is the price of keeping what you learn.

Let AI do the work you have already mastered. Never the work that is still making you smarter.

What is the one habit that matters most?

If you change one thing, change the order: do the hard part first, then bring AI in. Read the article before you ask for the summary. Draft the email before you ask for edits. Sketch your own answer to the problem before you ask the model for its. This single move is the difference between using AI to check your thinking and using it to skip your thinking.

There is a hint of why in the brain-imaging work covered in the first piece: the people who leaned on a chatbot from the first second showed the weakest engagement and could barely recall what they had produced. Doing your own pass first is not about pride. It is about being the kind of person who can still tell whether the model’s answer is any good.

How should you read when AI can summarize anything?

Read the source first, then use AI on top of it, not in place of it. A summary is a fine way to decide whether something is worth your time, and a terrible way to actually understand it. The understanding lives in the reading. Once you have read the thing, here is the move most people miss: ask the AI to quiz you on it. Have it ask you five questions about what you just read and check your answers. That flips the tool from doing the work to testing whether you did, and the testing is where the learning sets.

How do you write with AI without losing your voice?

Write the ugly first draft yourself. It can be bad. It can be bullet points. The act of getting your own thinking onto the page is the part that builds the skill and produces something only you could have written. Then bring AI in as an editor and a challenger: tighten this, find the weak argument, tell me what I am missing, play the reader who disagrees. What you should not do is paste a prompt and ship whatever comes back. A reader can feel the difference between writing that came from a person and text that came from a machine, and your voice is the one thing the model cannot supply. It comes from having read widely and thought for yourself, which is exactly the muscle these habits protect.

Can AI actually make you sharper?

This is the part nobody tells you, and it is the best news here. Used the right way, AI is one of the finest tutors ever built. The key is to make it test you rather than feed you. Decades of learning research point at the same thing: pulling an answer out of your own memory builds far stronger recall than reading it again. In one well-known experiment by Henry Roediger and Jeffrey Karpicke, students who kept quizzing themselves remembered about 80% of the material a week later. Those who simply reread it remembered about 36%. Rereading feels productive and mostly builds the illusion of knowing.

So aim the model at that. A few prompts that turn AI into a tutor instead of a crutch:

• “Quiz me on this. Ask one question at a time and wait for my answer before the next.”
• “I am going to explain this back to you. Tell me what I got wrong or left out.” (the test of real understanding is whether you can teach it)
• “Argue the opposite of my position as well as you can, so I can find the holes in mine.”
• “Do not give me the answer yet. Ask me what I think first, then push on it.”

Same tool, opposite outcome. One version does your thinking. This version makes you do more of it, better.

Can you make AI do this by default?

You do not have to type those prompts every time. The major AI tools now let you set a question-first style once, in their settings, so the assistant quizzes and challenges you by default. The clearest sign of how much this approach matters is that xAI built it straight into Grok. In Grok’s Customize panel, alongside Concise, Formal, and Comprehensive, there is a preset called Socratic that makes the model guide you with questions instead of handing over the answer.

You can get the same behavior in almost any assistant. ChatGPT, Claude, and Gemini all have a custom-instructions or personalization setting where you can write something like: “Before answering, ask me what I think. Quiz me. Make me explain ideas back to you.” Save it once and it shapes every conversation from then on.

And it does not have to be global. If you only want the Socratic treatment for studying or one piece of work, set it on a single project instead, using Claude Projects, ChatGPT projects, or the equivalent, and leave your everyday chats normal. The good habit becomes a setting you switch on exactly where you want it.

How do you protect your attention?

None of the above works if you cannot hold still long enough to do it, and attention is the part most under attack. The evidence piece covered the research: the mere presence of your phone drains mental capacity, and average attention on a screen has collapsed to under a minute. So treat focus as the foundation it is. Put the phone in another room, not face down on the desk. Read one long thing a day without switching tabs. Do single-screen work in stretches, with the chat closed until you actually need it. You are not training willpower so much as removing the thing that keeps pulling you off the task.

What does a simple weekly routine look like?

You do not need all of this at once. Start with a routine this light and build from there:

That is the personal version. The same principles change shape when the people involved are children still building these skills, or a team where AI is reshaping who learns the job. Those are the next two pieces in this series. For now, the headline is simple and hopeful: you can keep the tool and keep your edge. You just have to stay the one doing the thinking.

Common questions

Isn’t it slower to do the work myself first?

Yes, in the moment, and that is the feature, not the bug. The learning research is clear that the methods that feel slower while you do them are the ones that produce durable skill and memory. Easy and fast usually means it did not stick.

Which AI tool is best for this?

It barely matters. The habit decides the outcome far more than the brand. Any capable assistant can quiz you or challenge a draft. If you want a starting place, see our guide to using Claude, but the same prompts work anywhere.

Is it ever fine to just take the answer?

Of course. Offloading a task you have no interest in mastering is what tools are for. The line is simple: offload the chores, keep the thinking you actually want to own.

What if I keep slipping back into old habits?

Pick one habit, not five. The order rule, do the thinking first, is the one that carries the most weight on its own. Get that automatic, then add another.

Sources

• Bjork & Bjork (2011): “desirable difficulties” and durable learning (UCLA)
• The testing effect: retrieval practice beats rereading (Karpicke & Roediger), via Georgia Tech
• MIT Media Lab: “Your Brain on ChatGPT” (2025 preprint, treat as preliminary)

You may also like

• The Tool That Uses You (part one)
• The Reading Crisis Before AI (part two)
• What AI Literacy Means
• Physical Books, Handwriting, and Learning
• Glossary: What Is Cognitive Offloading?
• How to Use Claude AI
• The full AI and the Human Mind series
• The daily Beginners in AI newsletter