At the start of this year, my students asked whether they could use AI on the first brief: a weather app inspired by an artist of their choice. I said no. Not because I am against it, and not forever, but because it was the first brief of the year, and the first brief is where I find out how someone thinks. When I explained why, some of them nodded and some of them clearly didn't. Both reactions were fair. A brief like that is almost made for a machine: name an artist, describe a style, and something plausible appears in seconds. Which is exactly the problem. The interesting part was never the style. It was deciding how a way of seeing the world becomes a forecast someone can read at seven in the morning. That decision is the brief, and I wanted them to make it with their own hands at least once. It also left me with a question I have been carrying around for months, and it usually arrives at the end of a long day of tutorials: how do I teach you design if you don't design? Not "if you don't use Figma," or "if you skip the sketchbook." I mean the act itself. If the screen arrives finished, generated from a paragraph, where exactly does the learning happen?
For most of my teaching life, I have relied on something I rarely had to name. When a student draws a screen by hand, or drags a box across a Figma frame, every move is a tiny decision, and every decision answers back. The line lands a little too close to the edge. The button looks heavier than intended. Something that seemed obvious in your head looks wrong on the page, and you adjust. Donald Schön called this a "reflective conversation with the situation": you make a move, the move shows you something you didn't expect, and you respond. It happens dozens of times an hour, mostly without anyone noticing, and it is where judgment is actually built. That, in the end, is why I said no to that first brief. Not to protect students from a tool, but to make sure they had that conversation at least once before something else started having it for them.
The critique that matters is not the one at the end. It is the one that happens while you make.
A prompt skips that conversation. You describe, the machine decides, and you receive an answer that is already resolved. The critique still happens, but it arrives at the end, all at once, and it tends to sound like a customer rather than a maker: I like it, I don't, make it more something. It's not quite thinking about design. Not quite thinking with design. It is reviewing a delivery. I am not the only one noticing this. A 2026 study from the University of Melbourne compared sketching and text prompts as ways of working with AI image tools. The participants clearly preferred typing, even though the researchers argue that sketching better supports exactly this kind of reflection-in-action. A review in Trends in Cognitive Sciences this summer makes a related point more bluntly: when we hand thinking to AI, we skip the practice that builds specific skills, and those skills fade. Tools that offer hints instead of answers protect the learning. Tools that offer answers, it turns out, offer answers.
The problem with a finished answer is that it leaves nothing to argue with while you make it.
I don't want to pretend the answer is to ban the tools. "No" worked for one brief. It will not work for a career, and my students will probably work with AI for the rest of theirs. The real question is a teaching one: if the pencil was where the conversation happened, what is the equivalent now? Where do I put the small, repeated moment where your own decision looks back at you? My current answer is that the conversation has to be moved, not abandoned. Before the prompt, a rough sketch, so there is something of yours to compare against. After the result, an annotation: what did the machine decide that you didn't? And instead of a better prompt, a list of decisions, written in your own words, that you would defend in a crit. None of this is quite drawing. All of it is designed to make you argue with the work while it is still forming.
Margin note: The next time you generate a screen, circle three decisions on it that you didn't make. Then decide, for each one, whether you would have.