In 2018 I wrote a master's thesis called I, Designer. Fifty-odd pages asking whether design jobs would survive the rise of machine intelligence. I was in my twenties, finishing a degree, trying to take seriously something most people around me were still treating as sci-fi at best, paranoia at worst.
[You can read the full thesis here it's in French, it's very 2018, and I stand behind most of it.]
Seven years later I want to revisit it. Not to pat myself on the back for being prescient. Not to apologize for what I got wrong either. Just because the questions it asked have answers now, partial and uncomfortable ones, and I think they're worth sitting with.
What I actually argued
The thesis had two parts. The first tried to figure out what design actually is, stripped of the buzzwords and the LinkedIn superlatives and the "I solve problems not pixels" clichés everyone was saying at the time. I landed on something I still believe: design lives in the gap between what someone asked for and what they actually need. Not aesthetics. Not tooling. The gap.
The second part asked what happens to that discipline when machines get good at it.
My conclusion in 2018 was cautiously optimistic. The designer of the future, I argued, would be less of an executor and more of a director. Low-skill tasks would migrate to machines. The interesting work (the translation work, the judgment work, the work of deciding what the brief should have said) would stay human. The designer as bridge between the client and the machine, bringing the technical knowledge and the human taste the algorithm couldn't have.
I thought this transition would take longer. I thought we'd have more warning. We didn't, but I'll get to that.
What the market did instead
Here's what I failed to account for: the ceiling.
When I wrote about machines taking over repetitive design tasks, I put the ceiling of automation somewhere around "generate a layout from a template." What actually happened is that the ceiling kept moving, and it moved fast, and it did not politely stop at a level that preserved the interesting parts of the job.
Midjourney produces brand moodboards now. Figma's AI suggests component structures. You can paste a brief into a chat window and get a landing page that is, by most clients' standards, good enough. What I called "low-skill" in 2018 turned out to include a pretty significant chunk of what mid-level designers were billing for. That's the thing nobody wanted to say out loud.
I also spent a whole chapter on the crowdsourcing threat: Wilogo, Fiverr, the race to the bottom. That looks almost quaint in hindsight. That was disruption from below, humans undercutting other humans. What happened instead was disruption from above, tools that just absorbed the category entirely.
The clients didn't just find cheaper designers. A lot of them stopped hiring designers at all.
What I got right, which doesn't entirely comfort me
The conclusion I wrote in 2018 (designer as bridge, designer as director, delegation of low-level tasks to machines) is, in broad strokes, what actually happened to the practitioners who survived and thrived.
The designers doing well right now are not the ones who got faster at Figma. They're the ones who got better at reading a room. At translating a client's incoherent brief into something a product team can actually execute. At standing in a meeting between the person who commissioned the work and the machine that built it, and saying: "this is technically correct and it's also wrong, here's why."
That's the job. I described it in 2018 as the job of the future. The uncomfortable truth is it became the job of the present faster than most people were ready for, and a lot of people got left behind in the transition.
I watched some of it happen. I lived part of it.
The question the thesis didn't ask loudly enough
There's a section in the original about what I called the designer's "portfolio of competencies": the expectation that designers would need to constantly expand their skill set, know a bit of code, a bit of marketing, a bit of data. I framed it as market pressure. A constraint imposed by budget-conscious employers who wanted one person to do the work of three.
What I missed: that pressure was also a survival mechanism.
The designers who had built that breadth (who could talk to a developer, who understood what an API was, who had strong enough opinions to push back on a product manager) were the ones who became irreplaceable when the tools commoditized the production layer. The thesis treated polyvalence as a burden. Looking back, it was armor.
On AI as upstream, not downstream
The framing I used in 2018 positioned AI as a downstream tool, something that processed what designers created, that came after the decision. Adobe Sensei integrations, auto-suggest layouts, smart resizing. Helpful, a little unsettling, but still assistive.
What changed is that AI moved upstream.
A client can generate a first draft before they've even talked to a designer. A product manager can mock up a flow in an afternoon. A startup can launch with a visual identity that came from a prompt, and the market often can't tell the difference, and sometimes doesn't care.
The threat in 2018: the machine might replace you at execution.
The threat now: the machine might replace the conversation that used to precede you.
If the brief can be written, the moodboard assembled, and the first draft generated without a designer in the room, then what the designer brings has to be something that can only happen with a designer in the room. Judgment. Advocacy. The willingness to say "this isn't the right answer to the right question."
That's a harder thing to sell. It's also, I think, the only thing worth selling.
What I'd write differently
I'd be less optimistic about the timeline, obviously.
I'd be more honest about the economic pressure on mid-level practitioners specifically, not the senior designers with strong opinions and established client relationships, not the students still in school, but the large middle population doing competent production work who had the least runway when the tools changed.
I'd also write more about what I've started calling "the transmission problem": the fact that senior designers became senior by doing work that is now being skipped. The instincts that let a creative director catch a bad brief in sixty seconds were built by ten years of writing bad briefs and having someone explain why. If that experience layer disappears because AI handles it, we're quietly building a debt in human expertise that won't appear on any balance sheet until it's too late to pay back.
And I'd end differently. Not with the designer-as-bridge metaphor, which is right as far as it goes, but with something more specific.
The designers who will matter in the next ten years aren't the ones who resisted the tools, and they're not the ones who surrendered to them either. They did something harder : they built a relationship with the machine that preserved their own judgment. They know what to ask for. They know when the output is wrong. They know why it's wrong. And they can explain it to someone who never learned to ask any of those questions.
That's not really a design skill or a technical skill. It's something closer to literacy, in the oldest sense : the ability to read what a system is telling you and understand that it isn't always telling you the truth.
In 2018 I thought we'd have more time to develop that literacy before we needed it.
We didn't. And here we are.
Seven years on from the thesis, I'm in the middle of learning to code, a story I wrote about separately if you want the longer version. If you're navigating any version of this transition, from design or otherwise, I'd genuinely like to know how you're thinking about it.
