Buzz Usborne has spent 22 years in design and is Principal Designer at Help Scout. This year he rebuilt the Help Scout Design System so AI could read it, then wrote up how in Designing with AI. We last spoke five years ago, about remote work. This call was booked for 30 minutes and ran for an hour: on the two kinds of company AI is making, who gets to build now that anyone can, why he still starts on paper, and the old wooden desk he wrote into Help Scout's design system.
Design feels like it's in the middle of a transition. Where are we in it right now?
Someone said recently that we're in the messy middle, and I like that. It feels more organised than six months ago, but still messy. And in the messy middle, everything looks like failure. That's certainly where I'm at right now.
That's the coaching I give a lot of designers as well: experiment with everything, allow yourself to be excited by possibility, but don't get too attached to any particular way of working just yet. We're still figuring out what works, and what feels revolutionary today might be obsolete in six months. Enjoy the play!
I've been talking to a lot of people across the industry who've been through major technological shifts before: experienced designers and leaders at places like Mixpanel, Clay, Intuit and Shopify. I've been curious how they're approaching this transition, what's working and what isn't. And what's reassuring is hearing "Nothing's working." Despite all their experience, nobody has a definitive playbook for this yet.
That's more reassuring than going on LinkedIn and reading "Design is cooked." That doesn't help designers. So I've taken to the socials to say, "No, I haven't got a clue. No one's got a clue. You can relax."
For years the rule was never to jump straight to the solution. Now everyone is rushing to it. What happened to discovery?
AI dramatically increases our capacity to execute, but it doesn't give us more time, attention or better judgement. We can suddenly take on far more than before, but that just makes the decisions about what deserves our attention even more important. The bottleneck is shifting from making anything to deciding what deserves to be made.
I'm seeing a lot of people jumping straight from identifying a problem to building a solution, almost skipping discovery altogether. The quickness is attractive, but anyone who's designed long enough knows to be suspicious of their first idea. It's rarely the best one. But when AI makes it possible to go from an idea to a working feature in an afternoon, there's a temptation to treat that first result as the finished product. "I spoke to a customer and they had a request, so I got Claude to build it, and now it's live." That's exciting, but it overlooks a huge part of what design actually is and might actually cause far more problems than it solves.
I'm seeing two paths companies are taking at the moment. Some are primarily using AI to reduce the cost of execution: fewer people, less process, faster output. Others see the same efficiency gains as an opportunity to raise the quality and ambition of their ideas. Both are moving faster, but one is focused on making the existing work cheaper, while the other is asking what's now possible that wasn't possible before. I think that distinction is going to matter enormously.
My organisation understands design, but even they are saying, "We need to ship 10 or 100 times faster than we were before." Then it's hard to say, "Can we slow down a second, just for this one thing?" That has become the challenge for me as a design leader: figuring out how to move much faster, and how to earn the right to go slow when it really matters.
If everyone ships that fast, won't users notice the difference, and leave for whoever actually cares?
I had a conversation with a big company a few weeks ago, and they said something that's been on my mind since: the biggest competitor to anyone's product is your customers vibe-coding it over the weekend. Completely replacing what you do with their own bespoke version.
I think the differentiator to products worth choosing is going to be the accumulated decisions behind it. The coherence of the experience, the trust you've built, the tactility, the personality. And importantly, all the things you've deliberately chosen not to do. Great design is as much about what you leave out as what you put in. That's much harder to replicate, and I think that's where differentiation will come from.
One of the defining videos I've watched this year was Jony Ive talking to Patrick Collison at Stripe. He said people don't really recognise good design versus average, but they can feel it. When you use something you feel was created for you, you build an affinity for that product. That's what we felt 10, 15 years ago, when I used to work for these indie companies that had a cult following. Think Big Cartel, or Campaign Monitor and Mailchimp. People lived and breathed those brands because they could feel the care and attention that had gone into them. I think that's coming back.
The biggest competitor to anyone's product is your customers vibe-coding it over the weekend.
I've become lowkey obsessed with Unreasonable Hospitality by Will Guidara, whose philosophy of hospitality also influenced The Bear. There's a scene where a chef at a Michelin-starred restaurant is painstakingly peeling mushrooms. Someone questions why she'd bother since there's no practical benefit to it. But she explains that when someone notices that level of care, they know somebody has gone out of their way for them.
I love that. It's such a simple expression of what great design should feel like.
That's what we're doing here. We're trying to connect to people through the screen, and we don't do that by throwing stuff at them and showing them more. There's more content being produced by AI than ever before. I'm not sure whether the images I look at are real anymore, or if the news is real, or if a phone call is real. So I want to feel something. That's the side of history I want to be on: creating meaningful human connections instead of just proliferating more.
In Designing with AI, the breakthrough came when you skipped Figma and went straight to HSDS Studio. Who gets to use it: everyone, or only designers?
Anyone can use it. We've seen non-technical people make some genuinely thoughtful improvements and bug fixes, which has been awesome. But it's also created a bit of tension now that anyone can build features that look and feel like the real product without involving design. And despite best intentions, there's a risk of those changes reaching production without proper design oversight. That's uncomfortable. I don't love that.
The interesting challenge is that we can no longer be the gatekeepers of what gets shipped. And I don't think we should be. I want anyone in the organisation to be able to explore an idea, build a prototype and understand the implications of what they're proposing. But that doesn't mean that anyone should be shipping work without thorough review. Our responsibility is to make the design process accessible without sacrificing the coherence and quality of the product.
I specifically said to the design teams I work with, "We cannot be a bottleneck." I feel like our jobs depend on us not being one. We can't say, "We can't change our tools, we can't adapt our processes, we want more time." Or hide behind taste and feel, which all sounds too emotional when the industry is saying, "Ship, ship, ship."
So we built this tool to say, "You can build features now. Let's not put them into production, but you can get a feel for them." And it's been pretty good, because it helps anybody realise that design is hard. Someone says, "We just want to add this one button to the editor," and they add it in Claude, and then they say, "Oh, that's weird, that button looks like that button over there, and now they're competing for space." They get to see it. And those people can go away and say, "Do you know what? It's not worth it."
The other side is that HSDS Studio has given designers the ability to take on way more than we could ever have taken on before. All the things we punted in the past for scope's sake, we can go back and add now. And it's always those things that are the most delightful. That was the burden of design. We'd come up with great solutions to tough problems, and the first exercise was, "What do we cut?" And what we ended up cutting was the delight, the stuff that really connected humans. So now it's our responsibility to go back and add all that stuff back in.
Is there a rulebook people learn first, or do they go at it and you review afterwards?
It's a bit of both. I'm still working on an art director kind of skill. You can say, "Hey, evaluate this prototype," and it runs it against our design.md and our design principles and says, "We're asking a lot of trust from the customer without demonstrating much value." But I feel like people aren't using it, because prototyping is more like, "I want something done in a minute, and I want the gratification of it being done."
Help Scout is like an old wooden desk in the front room of your house, with afternoon light, a fresh cup of coffee and a blank piece of paper.
I wrote the design.md by hand. The design generators kept producing the same-looking thing again and again. It was just Tailwind.
I've also started writing an emotional tone for how our product feels. I was writing about how Help Scout is like an old wooden desk in the front room of your house, with afternoon light, a fresh cup of coffee and a blank piece of paper. Trying to evoke a feeling, because that's what these agents need. Not just a functional representation of components, but how these things feel to a user. Ultimately, that's what defined good versus not acceptable in our design reviews: the work fell short of our emotional tone.
It's largely the agent reading those docs and interpreting them. And it's never producing anything right off its first pass. Never. I'd say it gets us 40 to 60% of the way there. Or at least it gets us to a point where we're like, "Oh man, that is not worth taking on."
The fear I hear most from designers is losing the canvas. It's been the job since Paint and Photoshop. Do you still work in it?
The majority of my work still happens in the canvas. You develop a sense, as you become more senior, of what you can one-shot. But I find a lot of value in sketching. I still use pen and paper. I still go into the canvas and move ideas around, because I'm still a lot faster than Claude is. But I'm trying to find out where I sit on an idea.
My mentor always said a sketch is the fastest way from your brain, out of your arm, into your hands. I can't use Figma as fast as I can sketch a couple of boxes and go, "Well, that's not going to work." And then it's done.
Years ago I spoke about a creative scale of 1 to 10: 1 being the most obvious, beige, average solution to a problem, and 10 being completely insane. At agencies we'd ask customers, "How crazy do you want to go?" And it was always worth pursuing a 10, because you can always bring it back down. If you start with the average, you just produce the average.
AI is a good starting point and it's a good finisher, but it's not the middle. And the middle is where we do our best work.
That's even more important now because, left to its own devices, AI tends to produce things that feel generic or obvious. If you one-shot a feature and accept the first result, you're probably going to land somewhere around a one or two. Not because AI can't help produce original work, but because originality requires someone to deliberately push beyond the obvious. That's still our job if we're shooting for innovation.
AI is a good starting point and it's a good finisher, but it's not the middle. And the middle is where we do our best work. It's where we take a problem and say, "What can we remove from this? How can we do this so it feels like magic, or makes someone smile?" Those are the questions AI doesn't really know how to answer, because they're subjective. They're personal.
We designed a feature recently where we asked, "What if the cursor turned into a pizza while this thing was happening?" We could have lo-fi beats. We could give wacky prizes to the people who answer the most support queries. We never did any of that. But I love that we're messing around in that area, because that's what AI affords us. If we had to ship something this afternoon, we can do that. But let's spend the afternoon dreaming a bit, because now we can.
That's been the ultimate playbook for me. We use AI where we can to move really fast, and then we use the equity from moving fast to go really slow in the areas where we know it's going to be a differentiator. We can't just tell the organisation that quality takes time and expect everyone to accept that. We have to demonstrate that we can move quickly where speed makes sense, get things into customers' hands and solve real problems. But then use the time and credibility we've earned to go deliberately slowly in the places where thoughtfulness, originality and judgement can make a real difference. And while we're doing that, let's have this undercurrent of thoughtful, playful, weird, wonderful exploration.
Let's wrap up where we started: the messy middle.
I'm happy being in the messy middle as long as we all agree this is the messy middle. I've always been very cautious of anyone who takes to a stage and says, "This is how things are done." There isn't one way of doing this.
It's really messy right now, and that's kind of fun as well. I'm enjoying the unknown. But I also think AI can become an entropy machine. When everyone can create things at incredible speed, you can end up with more and more output that put together doesn't make a lot of sense. That makes shared direction, judgement and design leadership more important, not less. The faster we can move, the more intentional we have to be about where we're going.
It feels like the same conversations we were having when we moved from Photoshop to Figma, from desktop to mobile. It's all a bit chaotic and everyone feels like it's the end of the world. Eventually we'll find a way and this will all feel normal and sustainable again. That's how it goes.
Buzz Usborne is Principal Designer at Help Scout. His write-up of the design system rebuild is Designing with AI.