AI Makes You More of Whatever You Already Are

I’ve been in meetings where someone shares a document, and I start asking them questions about it. Why this framing? What’s the reasoning behind this number? And they don’t know. Not because they’re unprepared — because they didn’t actually write it. The AI wrote it, they skimmed it, and they forwarded it. The moment you push past the first paragraph and the conclusion, there’s nobody home.

I know the pattern well, because I used to do it myself.

In my early days with AI, I’d get an output that looked great, read the opening and the ending, decide both were solid, and ship the whole thing. Confidently. Then later I’d find something buried in the middle that I never would have signed off on if I’d actually read it. A claim I didn’t believe. A tone that wasn’t mine. A number I couldn’t back up. The AI hadn’t lied to me. I just hadn’t done my job.

That’s the whole subject of this guide: not whether to use AI, but how to use it without quietly getting worse at your own work.

The uncomfortable study

In 2025, a team at the MIT Media Lab ran an experiment that got a lot of people nervous. The paper is called “Your Brain on ChatGPT” (Kosmyna et al.). They put 54 people in three groups to write short essays: one group used ChatGPT, one used a search engine, one used nothing but their own head. Everyone wore an EEG cap measuring brain activity while they worked.

The ChatGPT group showed the weakest brain connectivity of the three. They felt the least ownership over what they’d written. And here’s the detail that stuck with me: many of them couldn’t quote a single line from an essay they had “written” minutes earlier. The researchers called what was building up “cognitive debt” — the mental version of skipping the work now and paying for it later.

Then they did the part that actually matters. In a fourth session, they switched people. The people who had been thinking on their own were now allowed to use AI. The people who had been leaning on AI now had to work without it.

The ones who’d built the thinking muscle first, then got AI? Their brain activity jumped. They recalled more. AI made them better. The ones who’d leaned on AI first, then had it taken away? They stayed checked out. Under-engaged. The habit of not thinking had set in.

So the finding isn’t “AI makes you dumber.” It’s more useful than that. The order you think in decides whether AI amplifies you or hollows you out.

A word of caution before anyone runs off quoting this as gospel: it’s a preprint that hasn’t been peer-reviewed yet, the sample was small, it was students in one city writing essays under a timer, and it only tested one chatbot. The lead researcher herself pushed back on the headlines — in her words, “we didn’t find any brain rot.” So don’t treat it as the final word. Treat it as a good, honest hint that lines up with something a lot of us have felt firsthand.

AI is an amplifier, not a character transplant

Here’s what I actually believe, and the study only sharpened it: AI won’t change who you are. It amplifies you.

If you’ve always looked for ways to do less — to slack off, to hand in something that’s technically finished — then AI is the most tempting shortcut ever invented. It’ll help you cut corners faster than you ever could alone. But if you’ve always had high standards, if you genuinely want your work to be good, AI won’t make you lazy. The opposite happens. You suddenly have a tool that lifts both the quality and the quantity of what you produce, and lets you attempt things you used to write off as beyond you.

Same tool. Two completely different outcomes. The variable isn’t the model. It’s you.

Which brings me to the one real danger, and it’s worth saying plainly. Using AI to do work faster is good. Using AI to avoid learning the work ends badly. If you reach for it specifically so you never have to understand the thing, you’ll produce output you can’t reproduce and can’t defend the moment the AI isn’t in front of you. That’s not leverage. That’s a loan against your own competence, and the study put a brain scan on the interest payments.

The sandwich, and why it’s only training wheels

The practical version of “think first, judge last” has a name some people use: the AI sandwich. You on the bottom, AI in the middle, you on top. Frame the problem yourself, let the AI generate, then evaluate what comes back before it goes anywhere.

I’ll be honest about this metaphor, because I don’t lean on it much myself. A sandwich sounds like exactly three moves — human, AI, human, done. Real work with AI almost never looks like that. Most of my best results come from a longer back-and-forth: I say something rough, it responds, I push back, we circle, it sharpens, I redirect. Closer to a conversation than a transaction. If you’re doing serious thinking, one round is rarely enough.

But as a starting rule for most people, the sandwich holds up well, because it protects the two things that actually matter no matter how many rounds you do:

You think before the first message. You decide what you’re trying to get and what a good result would even look like.

You judge before the last step. You’re the quality gate. Nothing goes out with your name on it that you haven’t actually read and stood behind.

Everything in the middle can be one exchange or twenty. The bread is non-negotiable. That’s the part people skip, and skipping it is exactly what the MIT group measured.

The front of the sandwich: define what “good” means

Most bad AI output traces back to a vague request. And after a couple of years of doing this fairly well, the mistake I still catch myself making is the laziest one: undefined adjectives.

“Build me an amazing website.” “Write me a strong strategy.” “Make this better.”

Amazing how? Strong by what measure? Better for whom? I haven’t told the AI anything it can aim at, so it guesses. Sometimes the guess lands and I get lucky. Sometimes it’s completely off. It’s a slot machine. Pull the lever, see what falls out. And the real cost goes past a bad first draft. When I never defined what good looked like, I have no yardstick to judge the result against at the end. I can’t be the quality gate on a standard I never set.

So the front of the sandwich is mostly one discipline: say what you actually want, and say how you’d know it’s right. “Here’s my situation. Here’s the tension I think matters. Give me three options, then argue against each one.” That’s a request the AI can chase and, just as importantly, a request I can grade.

If you don’t know how to frame it, there’s a shortcut that works embarrassingly well: make the AI do the framing with you. Ask it to interview you first — “before you answer, ask me five questions that would help you do this well.” It’ll surface the context you forgot you needed to give. Remember, it knows nothing about your specific world until you tell it.

The back of the sandwich: read the whole thing

The other half is simpler to describe and harder to actually do: read what came back. All of it. The middle included.

The test I use is embarrassingly low-tech. Could I defend every line of this if someone asked me about it, the way I’d want to defend my own work? If the answer is no, if there’s a sentence I’m hoping nobody probes, then it isn’t ready, and honestly it isn’t mine yet. That’s the moment to edit, push back, or cut, not to forward.

This is also where you have to be careful not to hand your judgment to the AI itself. It’ll happily tell you the draft is excellent, because these models tend to agree with you. Its praise is not the quality gate. You are. The AI can help you spot weak spots if you ask it to, but the sign-off is a human decision, every time.

When you can go light on all this

I don’t want to turn a useful habit into a guilt trip. If you’re using AI as an occasional helper — a quick reformat, a first pass at an email, brainstorming names for a dog — you don’t need to run a formal frame-and-evaluate ritual. Low stakes, low ceremony. That’s fine, and pretending otherwise would be exhausting.

The sandwich earns its keep the moment the work matters: anything going out under your name, anything someone will make a decision on, anything you’re supposed to understand and can’t fake later. That’s where thinking first and judging last stop being nice-to-haves and start being the difference between AI making you better and AI slowly making you worse.

The people getting radically more out of AI and the people quietly rotting behind it are, most days, using the exact same tool. What separates them is whether they still do the thinking on either side of it. Steter Tropfen höhlt den Stein — constant drops wear through stone. The habit you repeat is the one you become. Pick the one worth amplifying.

New to all this? Start with getting started with AI, then come back for the habit that keeps you sharp. Want to get genuinely good? Here’s how I’d actually do it.