AI didn't make you stupid. The interface did.

It reflects what you bring, they say. But who designed the mirror, and who did they design it for.

Back in university I took a course everyone loved to hate. It ran across the first three years of our program, and four times a year we’d meet in small groups to reflect on whatever course we’d just finished. The timing was always terrible. You’d just survived your calculus exam or handed in a big project, and instead of collapsing for a few days, you got to write about what you thought about the course and your own education. The audacity. (And shoutout to the three people from my class who subscribe to this newsletter. I’m pretty sure you also liked that course.)

Sometimes there was an extra assignment. Pick a book from a list, read it, write about it. The books were all about how humans and technology shape each other. The “tech is changing us” wave of popular nonfiction. Why habits stick, why screens make us lonely, how algorithms decide what we see. One year I read The Shallows by Nicholas Carr.

Hi, I'm Emma. Designer, engineer, and mom of two in Stockholm. I (over)think out loud about what it feels like to live with technology that thinks, and what our minds actually need to thrive.

Carr’s argument is that the internet is rewiring our brains for shallow, distracted thinking. He was so convinced of this that he moved to a cabin without internet to be able to finish writing the book. I was 23, studying media technology and human-computer interaction, and I wrote a long response arguing with his ideas. Yes, I noticed I couldn’t concentrate while reading his book (maybe it was boring?) Yes, I caught myself opening other apps when I picked up my phone to take notes. But his framing bothered me so much I still feel my body get ready for a debate when I think about the book.

Because if the brain adapts to what we do with it, why is that automatically bad? Why does he treat the digital generation’s new capabilities as lesser? What if we weren’t losing intelligence, but instead we were using it differently. I know how that sounds. People said the same thing about calculators. But nobody’s calculator was quietly deciding how they should approach a math problem. Or did it?

I dug up that paper recently. In it, I wrote that I wasn’t sure there was a right way to use digital tools. That maybe the answer was in designing them around different people’s actual needs. That the real question was why we keep designing one version of it and expecting every brain to fit. Cred to you, past Emma. I still believe we should adapt technology to people instead of adapt people to technology.

Around the same time, I read Clive Thompson’s Smarter Than You Think, which looked at the same evidence as Carr, the brain adapting, habits changing, attention shifting, and arrived at the opposite conclusion. Yes, technology is changing how we think. That’s what technology has always done. Writing changed how we think. The printing press changed how we think. The brain adapts. That’s not even a question anymore. The question is whether we treat the adaptation as loss or possibility. Thompson used a phrase that stuck somewhere in my brain and stayed there for over a decade: cognitive exoskeleton. The idea that technology doesn’t replace thinking, it extends it. When I started writing about AI some ten years later, I reached for those exact words without remembering where they came from.

I looked both authors up recently (or I asked Perplexity to, if that matters to you.) Carr argues that AI and the wider internet are accelerating our slide into shallow, distracted thinking, while Thompson talks about augmented cognition.

Does technology make us dumber or smarter? Well, it depends... on you.

I keep hearing that AI makes us dumber. I feel the opposite, more capable than ever. When I wrote that recently, almost every response said a version of the same thing.

It depends on how you use it. It’s a mirror, it reflects what you bring. If you’re smart, it makes you smarter. If you’re lazy, it makes you dumber. It’s a multiplier. It’s a tool, and tools are neutral.

They’re probably right, but they’re all saying the same thing: it’s up to you. You, the individual, are responsible for whether AI makes you better or worse. Even if you believe tools are “neutral” in intent, they’re never neutral in their effects, because design choices quietly tilt you toward some behaviors and away from others. Use it well and you’ll thrive. Use it badly and you’ll decline. And if you feel your capacity expanding, you were probably smart to begin with. Thank you for the compliment, but I refuse to let that be the whole truth.

Whether you think AI is a threat or an opportunity, the whole debate lands in the same place: the individual. Protect yourself, use it wisely, be intentional.

Meanwhile the interface just sits there, making choices for you.

Because every interface creates conditions. Those conditions make certain behaviors easier and others harder. They reward certain kinds of thinking and punish others. Not necessarily with error messages, just with friction. With defaults that assume things about you that may not be true.

Here’s what every AI chat interface in the world says when you open it:

ChatGPT: “What can I help you with?”
Claude: “What can I help you with today?”
Gemini: “Where should we start?”
Perplexity: “Ask anything.”
Grok: I genuinely don’t know and I’d like to keep it that way.

Every single one effectively assumes you’re starting from zero, that you already know what you need and can say it clearly. You have a text box no taller than five lines to do it in.

Sure, they can add a quirky greeting but the blank text input stares back at you either way.

That’s a design choice. And it probably works a lot better for a certain kind of brain. A leads to B leads to C. You have a question, you type it, you get an answer. Close the tab. That’s the kind of behavior it encourages. Pick your topic, have a conversation, if you get side-tracked or get a new idea mid-sentence you should open a new conversation.

Most of the time when I open the app, I’m not starting something new. I want to pick up a thread, or find something I said yesterday, or dump five half-formed thoughts and see which one has a shape. The thinking doesn’t happen before the conversation, the conversation is the thinking. I scroll through the vague auto-generated conversation names in search of the right conversation.

I can’t count the number of times I’ve asked Claude to help me locate a conversation that got lost. “I have this vague idea of a thing we talked about three days ago but I don’t remember what it was, just that it felt super interesting.” My prompting skills are only as good as my memory. But a bit of back and forth and we usually find it, because I’m not querying a system, I’m retracing a thought. And the only reason it works is that the model is competent enough to follow the associative retrieval my brain actually does. The interface won’t hold the thread, so I ask the conversation to be my memory instead.

The interface wasn’t designed for my use case. I just figured out how to work around it. I did what I usually do, overshare and overthink and called it thinking out loud because that’s a method in user research and it made it feel legitimate and not a waste of tokens.

And then it goes further. I built my own workarounds, little rituals and prompts that make the blank box behave, and so did everyone else, whole personal systems designed around exactly how one brain happens to work. The ones who figure something out package it into a course and sell it, a prompting method, a productivity system, a framework for thinking with AI. We each find what works for us, decide it generalizes, and sell it as the default. So instead of fixing the interface, we built an entire discipline around flattening our thinking to fit it.

It’s defaults all the way down. Even the critics land here. Jacob Ward argues in The Loop that we need to resist the feedback loops technology creates. Fight the algorithm. Protect yourself. It’s Carr all over again, just 10 years later. A decade of new evidence, same conclusion in the popular narrative: the problem is mostly framed as the technology, and the solution still lands on you to fix it, even when people like Ward also talk about companies and institutions changing their behavior.

But technology is created by people who are making decisions about what it is and how it’s supposed to be used. And once you start seeing the design and choices, you see it everywhere. You walk into a room and you notice the layout before you notice the conversation. You open an app and you see the choices someone made before you see the content. You hear a subway announcement and you think about who wrote it and why.

In Stockholm, the subway plays a recorded message: “If you’re wearing a backpack, please put it on the floor. Remember, your trip is also others’ trip.” That’s a system that acknowledges individual behavior happens inside shared conditions. It doesn’t hand you a backpack and trust you to be considerate. It nudges, and it designs for the outcome it wants.

Most current AI chat interface hands you a blank box and says figure it out. And when the outcome is shallow, the debate blames you. You should have prompted better, been more intentional, taken a course.

The model and infrastructure to hold a thread across sessions already exist.We have bigger context windows, longer memory, longer conversations, but all of it goes into making single sessions longer instead of holding the threads between them.

What I keep wishing for is smaller than a smarter model. I want a different front door. I open the app and instead of a blank box asking what I want, it shows me the threads still open in my thinking, the half-formed thing from Tuesday, the thought I almost had last night, and asks where I want to pick up. An interface that assumes I arrive mid-thought, because I always do. Or just something that adapts to me, or even better, something that’s invisible.

My 23-year-old self was already circling this in that paper, that maybe the answer was designing tools around the people who actually use them. The door I’d want is just mine. Yours would look like something else, and that’s the whole possibility, a door for every way a mind moves instead of one built for none of them. So before we ask again whether AI is making us sharper or duller, ask what your own front door would look like, if anyone had thought to build it for you.

---

If a specific interface came to mind while you were reading this, tell me which one and why. I’m collecting patterns.

---

I mentioned paid once or twice. Very quietly.

So here is me, finally saying it properly.

Here is a confession. Paid subscriptions have been on here for a while, and I have mentioned it. Quietly. A line at the bottom of a post, a soft note you could miss if you blinked. Never an actual ask. Which, if you know me, is an extremely on-brand way to ask for something without really asking. Put the option there, mention it at a volume nobody can quite hear, and half hope someone notices anyway.

A few of you did, and thank you, genuinely. But I never said it properly, out loud. So this is me saying it properly.

If something I have written has ever made you feel seen, or made you think for longer than a text box wanted you to, you can go paid. It’s the tip jar for a kind of writing the internet doesn’t reward, and it buys me the time to keep making it.

No pressure if it isn’t for you. I'll keep writing regardless, I don't really know how to stop. Paid just lets me do it with more time and less guilt.

— Emma