TL;DR
AI is not breaking the work. It is exposing what was already broken.
This is a counter to the class of AI commentary that gets rewarded for sounding worried without ever staking a position. Measured. Well-credentialed. Thoughtful. The genre collects social capital from depth-as-performance, while the cost gets paid downstream by everyone who still has to decide tomorrow morning.
The fear narrative is a choice. “AI can’t do the hard stuff” depends entirely on which AI you mean, and most people are criticising a tool nobody bothered to build properly. Two years at IKEA building exactly the thing that’s supposed to be impossible has taught me that. What is now being mourned was already being faked long before any language model arrived. And underneath all of it sits the real question, which is not what AI can do, but who you are if AI can do that.
This is not a technology conversation. It is an identity one. The disruption is happening. The only choice is whether you control how, or whether somebody else does.
Someone shared an article with me today that prompted a reflection (pun intended!).
It was written by Ethan Eismann, the CDO of Nubank, published on LinkedIn in late May. A sharp piece, well argued, drawing on the Designer Fund and Foundation Capital report on AI in design. Smart writer, smart company, real data. You should read it. It’s really good!
…And in the middle of it, there’s this sentence:
“what survives is the work AI can’t do (at least not yet).”
I sat with it for a few hours. Read it again. And realised that one sentence was doing more work in my head than the rest of the article combined.
Because the same observation, in nine different words, could (for example) read:
“what surfaces is the work we can’t yet do with AI.”
Same data. Same diagnosis. Opposite posture.
One frames designers as a species defending territory from a predator. The other frames AI as a flashlight, revealing the work that was always the real work. One is written from the cliff edge. The other is written from the doorway.
Language is never neutral. “Survives”, “can’t do”, “at least not yet”. Read those three phrases in a row and notice what happens in your chest. This article, like many others, doesn’t argue you into fear. It simply chooses words that already arrive there.
The reality is that most readers will not separate fact from posture. They’ll absorb the framing as gospel, file it under “what credible people think about AI”, and act accordingly. Which, in my experience, is how a profession (and the professionals within it!) ends up paralysed before the actual conversation has even started.
It is not just this one article, either. Over the last eighteen months, a particular class of writing has emerged around AI. Thoughtful. Well-credentialed. Measured. Sceptical in a way that signals depth without ever quite committing to a posture. The “genre” is rewarded with shares, head-nods, conference slots. It is the cool-kid version of staking a position, the one where you collect the social capital of seeming serious about the moment without having to make any actual call. The cost gets paid downstream, by everyone who still has to make a decision tomorrow morning. I work with those people every day. I am one of them!
The fear narrative is a choice
I’ve been in a version of this same conversation for about two years now. I’ve contacted first-hand with the people that face that threat in their day-to-day. Across nine countries, four continents. From junior to senior designers. Design leaders. Innovation teams. Researchers. Product people. Different markets, different cultures, identical conversation. The fear has the same texture everywhere.
One of them came back from a critique session at her company, frustrated. Designers with twenty years of experience each, dismissing AI work in the room without ever having opened the tool. I found myself saying to her, in the voice of the person nobody in that room had been willing to be: “I get it. You don’t understand it. And when people don’t understand things, they fear them the most.”
That wasn’t me being cruel. That was the thing somebody should have said, and didn’t. Those designers aren’t choosing fear because the fear is warranted. They are choosing it because curiosity requires them to admit they are falling behind. Fear is cheaper. Fear feels like wisdom. Fear, if you tilt your head slightly, looks like seniority.
Sticking your head in the sand and pretending the disruption isn’t happening doesn’t stop the disruption. It just means you will not be in control of how you get disrupted. Which may be the only thing actually worth being afraid of.
“Which AI?” matters more than you think
The article makes a claim I keep seeing repeated, almost word-for-word, in every other piece on the same subject: “AI accelerates the production of form. It doesn’t help you understand the underlying problem, and sometimes it actively obstructs that understanding by generating plausible answers before the question is sharp enough to be answered.”
I want to ask one question. Which AI?
Because out-of-the-box “ChatGPTs” are generalist consumer products. Brilliant ones. But also look nothing like “AI” as a capability. Confusing a chat interface with the underlying technology is roughly the equivalent of confusing a single Google search with the discipline of research.
For the last two years at IKEA, we’ve been building exactly the thing Eismann claims doesn’t exist. An agentic workflow for the innovation process. 15+ specialised agents that hand work to each other, each one shaped through context engineering, wrapped in our own framing, our own knowledge base, our own behaviour. Business case generation. Resource calculation. Cost structure. Virtual prototyping. Personas built on the fly, not as marketing posters but as interview subjects with enough cultural-context nuance. An orchestrator that runs the concept through an interview script that is not rule-based, but a live language model conversation. Tested. Operational. Keynoted at multiple industry conferences. Millions of euros on the table, validation underway.
Eismann is right. The crucial step occurs before all this. What companies tend to be genuinely bad at, and what costs the most, is the upstream work: figuring out whether we’re solving the right problem in the first place. Yes, we’re committing to building and testing faster and at a lower cost. But mostly earlier and way more often. In the problem space, not just in the solution space. Running synthetic research not to answer “will this work?” but to surface “what are we not even asking yet?”
One of my keynote slides reads: “We’ve been pushing AI to give us answers. What if its real value is helping us ask better questions?”
What that takes is not better prompts. It takes ownership. Depth of knowledge of the actual process we’re trying to amplify. Investment of money, time, and judgement to make the tool good enough to be trusted and adopted. Our agents speak IKEA better than any off-the-shelf model can, even if you ask it to. Because they have context. Because someone took the time to build it.
This is the missing nuance in almost every “AI can’t do the hard stuff” article I read. The hard stuff requires you to actually invest in making AI good enough to do it. Most people haven’t. So the AI they are criticising is the AI of someone who couldn’t care less about the specifics of your problem… of your company.
I say a version of this in almost every conversation I have on the subject. I invest my own money in testing tools. Not to be a hobbyist, but so that when I decide what is worth paying for monthly, that decision comes from somewhere. Informed by practice… not just theory. For that, you have to make the initial investment yourself. Otherwise, you don’t have a point of view. You just have an opinion. Which, I think, may be the difference between the people who will still be in this conversation a year from now and the people who won’t.
The thing nobody wants to say out loud
There is a quieter version of the fear narrative that I almost never see written down, and I think it is worth saying clearly.
A lot of people are not scared that AI is bad at design. They are scared because AI is becoming better than they are at the parts of design they were getting away with not being very good at.
For years, in many organisations, you could draft a persona on a Tuesday afternoon without speaking to a single real customer, present it on a Thursday, and nobody would check. You could run a workshop, harvest 150 post-its, cluster them into five themes, and call that insight. You could turn an interview transcript into a slide and call that analysis. The work, very often, is the performance of the work. Not because designers are lazy, but because the system rewards performance more than substance.
AI doesn’t just compete with that. AI exposes it.
If your insights workflow was already a slightly dressed-up transcript summary, then yes, of course a language model can replace it. The reason it can replace it is not that the language model is so clever. It’s just that the workflow was never very thick to begin with.
This is what I mean by the broken-before-AI problem. Personas drifted away from real customers long before AI showed up. Discovery work has been performative for two decades. The middle work that people are now mourning the loss of, the apprenticeship layer where juniors used to be made into seniors, was already being hollowed out in most companies, in the name of efficiency, way before any large language model arrived. When you automate something that was already broken, you don’t break it more. You just make a hairline fracture look like a wide open crack… make the breakage harder to ignore.
This is not a technology conversation
Every time I find myself in one of these debates, I notice the same thing. Almost nobody actually wants to talk about technology… or knows how to, really. The technology is just a proxy… and AI, the latest scapegoat. The true culprit is… well… people, systems, processes, old habits… fear of change! And pretty much the compounded circumstances that brought us here through the avoidance of accountability for its consequences.
What we are really talking about is identity.
Designers want to know if they are still designers. Engineers want to know if they are still engineers. Researchers want to know if their craft still matters. Strategists want to know if a thirty-page deck still has a place in the world. The question is never really “what can AI do?” The question is, “Who am I if AI can do that?”
That question is uncomfortable. It is also not new.
My background is in graphic design. If I had stayed there and become a super-specialist, I would be unemployable by now. I’ve lived through five of those reinventions since. The pattern is always the same.
There is resistance. There is dismissal. There is grief, often disguised as principled critique. There are conferences full of people (like me!) explaining why this one is different. And then, eventually, the people who took ownership of the disruption end up shaping it, and the people who insisted it was not happening get to live with the version somebody else built.
This is happening. With you, or without you. AI (or the world, for that matter…) won’t slow down for you to catch up later.
Where this leaves us
I’m not asking anyone to be optimistic. Optimism is a personality trait. I’m asking something slightly harder: be honest about which posture you’ve chosen, and notice that you chose it.
If the work you’ve been doing was always more about producing form than understanding problems, then “what surfaces” should land in your stomach. It should! It’s not a comfortable sentence. The work that surfaces is the work that requires depth, judgement, ownership, taste. In many organisations, that’s the work to be sacrificed first when budgets get tight. I know… I’ve been there. I’ve seen it first-hand.
That work was always the real work. AI is not erasing it. AI is putting a light on the room and quietly asking who was actually doing it.
The question I’d leave you with is not whether your job will survive. The question is: what becomes visible about your practice, your team, your company when the parts that could always be faked stop being part of the price?
If you take ownership of that, you are leading the disruption. If you don’t, you are exercising resistance against a train that is not stopping for you.
Pick.
And, yes, in case you were wondering I used AI to help me write this. Why wouldn’t I?
Which AI? My AI. The one I built. Trained on my methodology, my non-native English speaking voice, my work. I do the thinking. I do the writing. Then I run the draft through a panel of synthetic readers built after the kind of people whose judgement I respect (and that will show me no mercy!), to stress-test what I am about to say. AI doesn’t make the argument. AI made sure I made it cleanly.
Does it make this reflection better, or worse? Does it cheapen the thinking? How much of it is me, and how much is the machine? Is any of it even true? Does it matter?
Worth sitting with, these. They’re the questions every piece on this subject deserves. The difference, I think, is that I’m asking them out loud. Not just towards you, but myself, first.