We need your brain
Last week I was working through a challenging design problem. I needed to create a new set of screens for an app. Because of some limits on the hardware used for the app, there's a lot of constraints on how the problem can be approached.
The designs are complex with a lot going on, and because we need the app to work for non-technical users, the end result must be simple.
I first tried out Claude Code, running the ideas I had through it, and waited for results.
What came back was not good, at all.
Claude spat out less than mediocre concepts, breaking all conventions for best practice interfaces, and struggled to design anything coherent. It makes sense in hindsight because the constraints I gave it would have been hard for anyone to figure out.
The fault was mine.
I'd relegated my thinking to the machine, wanting to see what it'd give me. Sometimes this works; like pulling a slot and hoping for three cherries, you might come up a winner by prompting the AI until it provides a solution.
But, after wrestling through it for a while, I finally threw up my hands and pulled out my iPad—this ancient device, an analog of times gone by, an outdated mode for inputting—and sketched out what I wanted.
In thirty seconds I had a concept, and I was able to move forward. I was unblocked (though I must say, far from having solved this specific problem—that's kind of the joy of design, understanding the challenging and tackling it heady on).
The task I'd given to Claude, to create an interface with specific constraints and with a bunch of necessary affordances—wasn't an easy one. It's not something that a junior designer could figure out quickly.
I've given years of my life wrestling through these exact ideas, and I have a proven pre-2026 process to work from idea to reality.
To expect an LLM to do the same is foolhardy. Tools like Codex and Claude are good at doing what you literally ask. But if you expect them to imagine the challenging, to realize the theoretical, you may or may not come out with a winner.
Sometimes the best solution is to dive in, do the hard thinking, and work through the solution with the help of tools—but not by offsetting your thinking to them entirely.
And so now, with fresh eyes after a good night's sleep, I've taken to the task again. I can choose whether to brute force with prompts, or I can decide whether to go through my tried-and-true design process that I've learned over the years.
The answer, unsurprisingly, will be somewhere in the middle. With my iPad charged and ready to go, I'm going to bring sketches in, critical thinking alongside, and try to figure out this problem with my experience and assisted by the right tools.
By stepping back, and sleeping on it, I've had a moment to consider the bigger question.
In truth the problem may not be worth solving, the answer might lie in not doing the work at all, reconsidering the user experience, making sure I'm asking the right question—all things that matter for a designer, things that should be considered outside of just putting pixels on a screen.
Oftentimes, when I'm trying to force a bunch of buttons and text to fit into a screen, I'll step back and ask why—try to understand the purpose of it all, and in doing that come through with a deeper understanding of the user and their needs.
An LLM can't do that.
They can look at their corpus of data and infer likely outcomes, but they can't get into the psyche of human reality and uncover why we make decisions.
I've been in a lot of user research interviews over the years. I'm always humbled and surprised at how people respond to designs. The thought that goes into creating things, all the attention given to affordances and interfaces, crumbles in the face of a person with limited time trying to reach a specific goal.
AI can't help me with that. That's on me to understand, to internalize, and to use to create the best designs possible.
In a world where we feel rushed to context switch, do more and faster with less thought, we all have a chance to slow down. We can look at problems critical, understand the why behind them, and decide what's worth doing—as opposed to diving in without thinking or consideration for the problem.
Note: It's unfortunate that I have to say this, but none of this was written with the help of AI. For English majors that's painfully obvious, I like my imperfect me, and though I strive to write clearly and usefully, I'd rather personality over mediocrity.