A prophecy landed on my desk
A few days ago someone sent me one of those long, confident predictions about the future of our industry. You've seen the genre: AI destroys design and development jobs, year by year, with dates and percentages. Frontend 70–80% automated by 2027. Software creation 100% democratized by 2029. The screen disappears by 2031.
It was well written. It was brutal. It was built to go viral.
And here's the uncomfortable part: half of it deserves to be taken seriously. The other half is science fiction wearing a spreadsheet.
I spend my days embedded in an agentic AI initiative inside one of the largest retailers in the world, designing how autonomous systems and humans share decisions worth real money. So this isn't theory for me. It's my Tuesday. Let me separate the signal from the theatre.
The half that's true
The core mechanism of the prophecy is correct, and pretending otherwise is cowardice.
AI doesn't take your job. It multiplies someone else's. The real threat was never a robot sitting in your chair. It's a senior designer or engineer who, armed with agents, produces what a team of three produced two years ago. Companies don't need the work to disappear to cut headcount — they just need output per person to double. It already has.
The junior freeze is real. Hiring for entry and mid-level roles in design and frontend has been quietly frozen across the industry. The tasks juniors used to learn on — production UI, layout variants, component documentation, boilerplate code — are exactly the tasks AI eats first. We are burning the bottom rungs of our own ladder, and nobody has designed the replacement.
Production UI is a commodity. Turning wireframes into high fidelity, generating variants, documenting a design system: this is no longer craft, it's throughput. If your value proposition is "I move pixels fast," you are competing with something that moves them instantly and never sleeps.
Mid-size agencies are getting squeezed. When a client can prototype internally in days, billing by the hour of production stops making sense. The consultancies that survive will sell judgment, not hands.
So far, the prophecy and I agree. Now for the fiction.
The half that's fiction
Fiction #1: the decimals. "Frontend will be 70–80% automated by 2027." Where do those numbers come from? Nobody has them. Not the labs building the models, not the analysts, certainly not a blog post. Generating code is not shipping software. Integration with legacy systems, security, compliance, maintenance, the archaeology of ten-year-old databases — that's where most of the cost of software actually lives, and it automates slowly and painfully. AI writes code at superhuman speed; making that code survive inside a real organization is still stubbornly human.
Fiction #2: "software creation democratized at 100%." I've heard this one before. No-code was going to democratize software in 2015. WYSIWYG editors were going to do it in 2005. Visual Basic was going to do it in the nineties. What actually happens when building gets cheaper is the opposite of extinction: we build more software, and the bottleneck moves from producing to deciding what deserves to exist. That decision — what to build, for whom, and why — is not a technical skill. It's the core of product design. The cheaper the execution, the more expensive the judgment.
Fiction #3: the death of the screen. This one is personal. Years ago I co-founded a voice-based social network, working alongside the Google Assistant team. Back then, the same prophets told us voice would kill the screen within three years. It didn't. Screens survived because they are the highest-bandwidth channel between a machine and a human brain — nothing transmits more information faster. What agents will kill is the operational screen: the dashboard you babysit, the form you fill for the tenth time. The visual interface as a concept isn't dying; it's being promoted from workspace to control tower.
When someone predicts your extinction with decimals, they're not forecasting. They're marketing.
What the prophecy conveniently forgets
Three things never appear in these viral predictions, and they change everything.
The layoffs weren't only AI. A significant part of the tech job destruction of 2023–2025 was a macro correction — post-pandemic over-hiring meeting rising interest rates — rebranded as an AI story because "we hired too many people" sounds worse in a press release than "AI efficiency." AI is accelerating a trend; it didn't invent it. If you calibrate your career on the marketing version of the layoffs, you'll aim at the wrong target.
Accountability doesn't automate. When an autonomous agent makes a purchasing decision worth millions, or denies someone a loan, or reroutes a supply chain, a human being signs for it. Someone has to design the system of permissions, visibility and reversibility that makes that signature possible. That "someone" is a role that barely has a name yet — and it looks suspiciously like a designer who understands systems.
Taste doesn't average. Models are, by construction, machines of the statistical middle. They produce the most probable output. Markets reward the improbable one. The gap between "correct" and "worth choosing" is where taste lives, and taste is trained the old way: years of shipping, failing, and looking closely.
Where the role actually goes
The UX role isn't facing extinction. It's facing relocation — and relocations kill the people who refuse to move.
For fifteen years the job was designing how people operate software. The new job is designing how software acts on people's behalf: what an agent may do alone, when it must ask, how its actions stay legible, how it fails without taking you down with it, and — above all — whether the thing should be built at all. Less choreography of screens, more constitution of behavior.
What transfers across that move is not your Figma velocity. It's the fundamentals: user psychology, information architecture, research discipline, the ability to ask what problem are we actually solving? before anyone opens a tool. Tools rot. Fundamentals compound.
How to make the jump
This is what I'm actually doing, not what sounds good in a keynote:
Learn the logic, skip the syntax. You don't need to become an engineer. You need to understand how software is built — data, states, APIs, deployments — well enough to direct AI that builds it. A designer who understands architecture can orchestrate a product end to end. One who doesn't is decorating someone else's decisions.
Use AI as a stack, not a toy. Drafting copy with a chatbot is 2023. Use it to analyze thousands of rows of user data, to build functional prototypes in hours, to stress-test your own logic. If your AI usage fits in a chat window, you're leaving most of the leverage on the table.
Move up the value chain, deliberately. Own the problem, not the deliverable. The deliverable is getting cheaper every quarter. The problem — understanding a business, a user, a market, and betting correctly on what to build — is getting more valuable at exactly the same rate.
In summary
The prophecy is right about the mechanism and wrong about the theatre. Yes, efficiency per person is exploding and teams will shrink. No, nobody knows the percentages, screens aren't dying on schedule, and "everyone will build software" has been three years away for thirty years.
The honest version fits in two sentences. The risk to your career isn't artificial intelligence; it's standing still while the role relocates. And the surest sign that a prediction is selling you something is that it comes with decimals.
The future doesn't come with percentages. It comes with homework.
If you're navigating this same relocation, my inbox is open: hello@mveiga.com