For most of my career, design meant shaping stable things. Interfaces, flows, and systems I could understand, refine, and ultimately hold still. Confidence came from understanding, polish and predictability. If I invested enough care into the craft, I could feel reasonably certain I was on the right path.
That’s no longer the material I find myself working with, and I don’t think I’m alone.
We’re no longer shaping systems with predictable boundaries. We’re shaping how they reason and respond. That widens our impact, but it also deepens it. The consequences of our decisions are harder to predict, yet they matter more than ever.
We’ve been here before
This shift didn’t come out of nowhere. We’ve designed probabilistic systems for years: feeds, recommendations, playlists. Those systems were dynamic, but they were dynamic in selection. The outcome wasn’t always predictable, but the components were known. Content was bounded, tagged, enumerable. Even when you couldn’t guarantee what a user would see next, you could still reason about the space of possible outputs.
Generative systems are meaningfully different. They’re dynamic in construction. Output isn’t selected from a predefined set, it’s synthesised in real time. Language, structure, tone, and meaning are assembled on the fly and the experience is often never the same twice.
Designers may not fully understand how a system assembles a response, but that behaviour is the experience. And because users interact with it directly, design responsibility becomes harder to reason about, not less important.
When doing everything “right” isn’t enough
Years of practice taught me what “good design” looked like. Beautiful UI, thoughtful UX, performance, usefulness, and user research as a grounding force. Talking to real people, observing real behaviour, and using that signal to build confidence in decisions. None of that stopped mattering. Users still care deeply about quality and clarity. But those signals alone no longer guarantee confidence.
When the experience is probabilistic, the confidence I used to build through craft doesn’t arrive the same way. You can do the research, apply the product thinking, refine the interaction, bring everything you’ve developed over years, and still get outcomes that surprise or disappoint you. The process still matters, but it no longer guarantees convergence. That was the moment something started to feel missing. The user was interacting with something I hadn’t explicitly designed, yet it was still my responsibility to shape their experience. That tension changed how the work felt.
How do you design for something that isn’t the same twice? How do you take responsibility for an experience you can’t fully inspect?
These questions aren’t theoretical, they show up in day-to-day decisions. Over time, I’ve found a few things that help me stay grounded in this space.
Intuition
Designers are trained to notice when things feel off. We interact with designed things every day and develop a sensitivity to friction, confusion, and misalignment long before we can articulate them clearly. With even a basic understanding of how these AI systems work, intuition still helps orient us. It points toward opportunity. It tells us where something might be fragile, promising, or wrong.
This is one of the places where designers still matter deeply. New tools like Figma Make, v0 and Claude Code have collapsed execution barriers. Roles are blurring. PMs can build, engineers can design and models can generate. But intuition is still a human skill, and it works on two levels. There’s trained perception, the ability to detect weak signals early, to sense friction or fragility before it’s fully visible. And there’s taste, the sense of what good feels like, what makes something feel authored rather than assembled. As these tools raise the baseline, output converges toward the average. Perception tells you something is off. Taste tells you what’s missing. Both are trained through experience, and in unstable systems, both become critical.
Intuition alone isn’t grounding, though, it’s directional.
Foundations
Staying oriented requires more than instinct. It requires going back to fundamentals.
That’s meant stripping away existing UX patterns and interface conventions, and looking beneath them at human behaviour, cognition, and the underlying problems we’re trying to solve. This shift has been essential in understanding where AI can be integrated meaningfully, rather than simply bolted on.
For years, we built layers of efficiency and abstraction because technology demanded it. We optimised around constraints. New capabilities create moments where those layers can be questioned. They let us ask what was previously impossible, or what problems were only approximated because we couldn’t go further.
This has really helped me re-engage with problem spaces in new ways and try and understand the human role in all of it, with and without AI.
The scientific method
Leaning on the scientific method has helped give a structure and framework to combat uncertainty. Form a hypothesis about what might help, build something tangible to test it, observe how the system behaves, and make a decision based on what actually happens. Then repeat. Prototyping has been the vehicle for this process.
Building something tangible early, something you can experience, share, and interrogate, makes the problem legible. You give the system something real to respond to, and its behaviour starts talking back. Prototypes can reveal where things break, where they surprise you, and where something feels wrong even if it technically works.
This can look like starting with a solution, and it some ways it is. But what you’re actually doing is creating a concrete instantiation that allows learning to happen. You shape the system, the system responds, and that response informs the next move. Hypotheses are refined or discarded not because they sounded good, but because the behaviour made them untenable.
In this sense, prototyping is a thinking substrate. Like pottery, you can’t understand the form without working the material. The act of making changes what’s possible.
Judgment
Where intuition senses and proposes, judgment is what holds the work accountable. It asks whether something actually works. Whether it’s grounded in real needs. Whether we should ship this, or whether we’re outrunning what’s ready. Judgment isn’t feel, it’s evaluation. And like intuition, it’s trained. But it’s trained through consequence, not exposure. You develop it by living with the results of your decisions, not just by noticing what others got wrong. It’s what prevents the kind of overshooting that builds something impressive but isn’t actually grounded in reality or useful to people.
The pressure to move fast is real, we all feel it. Overshooting often looks like promising capabilities that aren’t ready, framing futures as present, or shipping abstractions instead of meaningful experiences. The result is fatigue, in designers and in users. “AI everywhere” without clarity or trust.
Restraint matters. Aiming isn’t slowing down. It’s choosing not to outrun reality.
Aiming as the work
The target is moving. Standing still can feel safe, but it isn’t. On the other hand moving blindly creates chaos and risks losing trust. The space in between, where we aim, adjust, and learn without pretending to know, is where I’m spending my time now.
This Substack is a place to think out loud about how we do that. Not with certainty, but with intention.
The job hasn’t changed. We’re still here to help people, to make things clearer, more usable, and more humane. What’s changing is the material we’re working with. Learning how to move with it, without overshooting, is the work.


