Earlier this year I was ready to believe that design was in trouble.
Not the craft. The value of it. I'm the Head of Design at a business that moved early and hard on AI, and for a few months I genuinely couldn't explain what my team was for in a way that would survive contact with a sceptical executive. Everything I reached for sounded like something I'd read on LinkedIn.
So I got back on the tools. Not reviewing other people's work, but properly back into a squad alongside Product and Engineering, because I wanted to make better decisions for my designers and the only honest way to do that was to find out what the job actually feels like right now.
Three things happened over the following six months. Each one told me the same thing.
Speed was the strategy, until the questions came back
Around the same time, the business shifted hard towards speed. That's the promise of AI, after all. It should make you go faster. What came with it was a preference for experimentation, and for speed to customer value over almost anything else. My team, Product Design and Engineering both, shipped as fast as we could. Quality went to the wayside, and the framing that allowed it was reasonable enough on its face: if it doesn't work we'll pull it out, and if it works we'll come back and improve it.
Fast forward about six months. The same stakeholders who'd asked for speed started asking why the UX on some of these things felt a little off. Why things didn't look as premium, or as polished, or as refined as they should. We'd made one big loop. Speed, speed, speed, quality can wait, and then six months later, where's the quality, where's the rigour in this work.
Nobody in that loop was wrong, exactly. But when you optimise for speed you skip steps, and the steps you skip are usually the ones protecting the person at the other end.
Story 1: A prototype validates the idea, not the experience
You'll know this one. A Product Manager puts a prototype together, sends it over, and asks you to clean it up so the devs can build it.
That happened a few weeks into being back in a squad. We were converting an order page into a newer design language, a tidy-up mostly. Then the PM wanted to add a small piece of functionality to the form: a toggle that would change the form dynamically based on what the customer selected. Fine. It'll be quick. Let's just do it.
The prototype already existed, so the idea was already settled. That's the part worth sitting with. A prototype is the thing designers use to prove an idea has legs, and when anyone can produce one in a few minutes, it starts doing that job on its own. The idea gets validated by the artefact that was only ever meant to test it.
So I took it and started cleaning it up. And as I stepped through what it would actually feel like to read this form and fill it in, I realised the two options we were presenting were confusing. Genuinely hard to choose between. Not a styling problem. A can't-make-a-decision problem.
Your user is still a human. That reads as too obvious to write down, right up until you notice how much of the process no longer has a human anywhere in it. Somebody has to sit inside it and ask how an actual person experiences this. For now, that's still us.
Story 2: Two half-hour conversations beat the prompt I didn't write
We'd done a deal with another business, and the outcome of it needed a small presence in our interface. I sat down with the manager who owned the work, and before anything else we brought up the platform and just talked about it.
Things surfaced as we went. Bits of logic we'd need to cater for depending on who we were solving for. Existing logic about what could and couldn't be included in the platform already. The more we talked, the narrower the solution got.
Two half-hour conversations. By the end of the second one we'd scoped it down to two minor changes to the platform. That was the whole thing.
I could have taken the brief, dropped it into Claude and asked it to show me how to solve this in our platform. I'm fairly confident that would have taken longer. I'm not at all confident it would have hit the right brief.
The mark of a designer who operates well inside a business is being able to have that conversation. Not producing a solution fastest, but working out who you're solving for and what the smallest right thing to ship is. That's a back-and-forth between two people who know different halves of the problem. It doesn't compress into a prompt.
Story 3: Sometimes it isn't taste, it's the tooling
This is the one I got wrong for months.
On the other side of the business we push changes into our Marketing site through an MCP. We're also standing up a new geography, a US expansion we're running much like a new startup, where speed is the whole game and being beaten to market is the risk that actually matters.
Feedback kept coming back from stakeholders in that space. The pages don't look like what we expected. For a long time I filed that under taste. I'd talk to the designers, get them to pay more attention, look more closely, care harder about the detail. That was my read, and I held it for months.
Then the designer working on it left, and I had to step in and do it myself.
Here's what was actually happening. The MCP would build from a Figma design, but it would reach for components already sitting in the site's library, and those components wouldn't style the way we needed them to. When I tried to make the styling changes myself, it blocked me. Not enough access. Not enough permissions. So I spent my time finding workarounds, trying to get close enough that the page at least held together.
At five o'clock on a Friday I accepted I wasn't going to get there.
My designers hadn't been careless. They'd been working inside the limits of what the tool would let them do, and I'd spent months quietly attributing a tooling problem to their judgment. That's the part of this I'm least comfortable with.
Being the human in the equation isn't only about the user. Sometimes it's noticing that the process itself is wrong, and recognising that you're the one with the standing to change it.
AI only pays you back if the trade stays honest
We trade quality for speed constantly, and usually without saying so out loud. That week was the clearest version of the trade going badly. Hours spent fighting a tool for an output I never got, when the same tool with the right permissions would have had me finished in a fraction of the time. That isn't AI failing. That's a setup nobody had gone back and fixed.
You have agency over the setup. Work out what AI is genuinely good for in your context, notice where it's quietly costing more than it returns, and then go and change the conditions so it works for you instead of against you. Nobody else is going to do that on your behalf, and the tools certainly won't raise their hand and tell you.
The answer to the value of Designers is a hopeful one
At my lowest point on all this, I asked Claude what the value of design is. It gave me a good answer. The iterative process. A visual practice that's distinct and hard to replicate. All of it correct, all of it reasonable, and none of it did a thing for me.
The answer turned up later, and it came from the work. From stepping through a form as a person and finding it confusing. From two conversations that did what a prompt couldn't. From a week of fighting a tool and realising I'd blamed the wrong people.
The problems haven't gone anywhere. The shape of them shifts, the tooling shifts. What stays in the middle of it is a person who notices.
Don't hand that part over.