→ Recap of Part I: Designers Must Look to Demis Hassabis, Not Jony Ive
As the paradigm shifted in the early 1970s, design leaders like Jony Ive ushered in a new era of personal computing. Focusing on surfaces that made technology ubiquitous and desirable. But that era is ending, and the real design work has moved to the intelligence layer beneath those surfaces. Where Ive architected poetry for the eye, Demis Hassabis is architecting poetry for the mind. He represents the new kind of leader, design as a craft desperately needs to look to.
When the Interface Thinks
For most of computing history, designers have been the authors of the visible world. They shaped the screens we tap, the icons we recognize, the patterns we repeat until they feel second nature. From Xerox PARC to Apple's Human Interface Guidelines, nearly every interface worth designing has already been designed. The metaphors of interaction: the desktop, folders, windows, menu, tap, swipe, are now complete, endlessly iterated but rarely reinvented.
But we are entering a new frontier. For the first time, the interface itself thinks. And when it thinks, the rules change.
Today's most profound design problems don't live on the screen at all. They live in the invisible substrate beneath it: how models behave, how they reason, and how their decisions are constrained or liberated.
This layer, once the domain of researchers and engineers, is now where the future of design will need to be written. The things being shaped today are not pixels and layouts, but policies, training data, weights, prompts, tool use patterns, and memory architectures. These are the true materials of intelligence design.
If designers don't go deep into this layer, someone else will. And if that someone else is only a researcher optimizing for accuracy or a company optimizing for profit, the human experience will be decided without us. Designers will be pushed to the margins, relegated to decorating the (surface) output of systems they never helped shape. And that, will truly be unfortunate.
We can already see this playing out, and it's embarrassing.
Most so-called "AI products" today are not truly designed. They are old metaphors hastily dressed up with a chatbot. Open ChatGPT. Open Claude. Open Gemini, Perplexity, Copilot. Line them up side by side and try to tell them apart.
You can't. Because these are interfaces trying to do many things for many people at the same time. With Apps in ChatGPT now, this becomes even more true. The speed at which this technology is moving makes it impossible to keep up. Which is why we end up with the current state.
A blank text box. A glowing cursor. A few suggested prompts floating above like placeholder copy. Maybe a sidebar for chat history. Marginal differences in color palette or typeface, but no new paradigm. It's as if someone took the command line from 1985, smoothed the edges, added syntax highlighting, and declared victory. Throwing the onus back on the user with lazy copy, “Ask Me Anything…”
These are not breakthroughs in interaction paradigms. They are placeholders.
And the reason is damning: most designers don't yet understand how to design for intelligence. They treat the model as a black box. An oracle that speaks when summoned, inscrutable and untouchable. So they do what they know: wrap it in familiar patterns. Put a nice font on it. Make the corners round. Add a dark mode and a little motion.
But the parameters that actually shape the experience, the things that determine whether an AI feels helpful or frustrating, aligned or alien, live inside the black box:
Context windows: How much memory does it have? Does it forget you after one conversation, or remember across months?
System prompts: What invisible instructions shape its personality, its boundaries, its refusal conditions?
Retrieval chains: When does it search for information? What sources does it trust?
Tool orchestration: Can it take actions on your behalf? How does it decide when to escalate to a human?
Policy boundaries: What won't it do, and why?
These are the things that actually shape behavior. The new "pixels" of experience design. And designers are almost entirely absent from these conversations. All we're doing is tidying up the edges of a system we don't understand and can't control.
The great irony is that this moment holds unprecedented creative potential. When the interface itself thinks, entirely new kinds of experiences become possible. Experiences we haven't even imagined yet, because we're still trapped in the metaphors of the past.
Ask yourself: if a word processor were invented today, knowing what we know about AI, would it look like Microsoft Word with Copilot or Google Docs with Gemini tacked onto the sidebar?
Of course not.
And yet that's exactly what we're building. We're draping a modern brain in a 1980s desktop UI and calling it innovation.
This isn't design. It's cosplay.
And if designers who continue to polish surfaces don't go deeper, if we don't learn to shape the substrate of intelligence itself, we won't just be irrelevant. We'll be replaced.
The Inversion of Paradigms
As intelligence moves beneath the interface, something strange is happening.
The foundations of design, assumptions we've held for decades about how software works and who controls it, are beginning to invert.
Not evolve. Not iterate. Invert.
In the age of static interfaces, designers shaped nearly everything a user experienced. The layout. The hierarchy. The flow. The micro-interactions. Roughly 85% of what someone saw and touched was the direct product of a designer's deliberate choices.
But as AI systems take over more of the decision-making, that balance is collapsing with terrifying speed. Soon, designers may control less than 5% of the experience, with the remaining 95% dictated by a model's behavior: its reasoning, its policies, its hidden weights, its choices made in milliseconds without asking permission.
What we're witnessing isn't incremental change. It's a phase inversion. The kind of shift where every rule flips to its opposite, where the leaders of the old world become obstacles in the new one.
It's happening across five fronts. Each inversion, on its own, is disorienting. Together, they represent the collapse of the entire paradigm designers have built careers on.
1. Affordances Invert: From Navigation to Negotiation
Old world: Interfaces were deterministic. A designer in Silicon Valley decided where everything lived on your screen, and you navigated their decisions. Swipe here. Tap there. Follow the breadcrumb trail they laid out. The map was fixed.
New world: You don't navigate. You negotiate. You describe what you want, set boundaries, express values. The model interprets, reasons, and acts. The interface isn't a map anymore. It's a dialogue.
The shift: Design is no longer about arranging buttons or optimizing navigation trees. It's about shaping conversations of intent between humans and machines. Deciding when to clarify, when to challenge, when to comply.
The problem: Designers are almost completely absent from this work. The very space where meaning is constructed, where user intent meets machine reasoning, is the space where designers have the least influence.
2. Authorship Inverts: From Creation to Generation
Old world: Software produced predictable outputs. You knew exactly who made a PowerPoint deck, or someone you trusted did. The artifact was static. The author was clear.
New world: Agents generate outputs invisibly. They draft emails, compile reports, coordinate decisions across systems. What you see isn't what you or someone else made. It's what a model assembled on your behalf, stitched from sources you didn't choose, using logic you didn't write.
The shift: Design once focused on creation moments. The blank canvas, the "New Document" button. Now it must focus on generation moments: how do you review, approve, or override something you didn't create?
The problem: When outcomes are dynamically generated by a model, the designer's role becomes undefined. Whose vision are you designing for when the system itself is the primary author?
3. Failure Modes Invert: From Visible to Invisible
Old world: Interfaces failed loudly and visibly. A button broke. A form didn't submit. A page crashed. The error was obvious, and fixing it was straightforward.
New world: AI fails quietly and probabilistically. A hallucinated fact that sounds perfectly authoritative. A bias embedded in training data that surfaces in subtle ways. A privacy breach buried three steps deep in a chain of autonomous API calls.
The shift: Failures are no longer deterministic bugs you can reproduce and patch. They're stochastic risks that emerge from the model's behavior. Invisible until they cause harm.
The problem: Most design teams still optimize for surface-level friction (loading states, button placement), while the deeper systemic risks of bias, hallucination, misalignment etc. go unaddressed. We're polishing the railings on a ship whose hull could be cracked. But we won’t know, because we’re not thinking about it as designers.
4. Values Invert: From Delight to Alignment
Old world: The north star of design was delight. We asked: "Is it beautiful? Intuitive? Does it spark joy?" Success meant users loved the experience.
New world: A harder question looms: "Is it aligned?" Aligned to whose goals? Whose ethics? Which laws? Which cultural norms? Success means the system acts in accordance with human values. But which humans, and which values? What Altman believes or what Musk X’eets? Sure.
The shift: In the past, designers were the custodians of the human voice in systems built by engineers. They translated messy human needs into elegant interactions. But in a model-driven world, those values are encoded in training data, system prompts, and policy layers. Territories where designers are rarely present.
The problem: Delight without alignment isn't just inadequate. It's dangerous. A delightful interface can mask a system optimizing for the wrong goals entirely. What’s worse is these systems create the illusion of productivity. “Use AI. Do more. Produce more. Or get left behind.”
5. Creativity Inverts: From Universal to Exclusive
Old world: Design creativity was about composition. Arranging pixels into harmony. Crafting flows anyone could understand. The designer's skill made the experience accessible to all. Canva, Squarespace etc. democratized the best of design for the usual knowledge worker.
New world: Creativity comes from model behavior. A perfectly tuned prompt unlocks magical responses. A new drug or molecule. A well-designed retrieval chain transforms a dull chatbot into a brilliant collaborator. But this power is exclusive to the adept.
The shift: The medium is no longer the screen. The medium is the model itself. And only those fluent in its language (prompt engineering, context tuning, policy design) can unlock its creative potential far greater than the regular person using AI.
The problem: A small technical elite can coax wonders from these systems. The average user? They fumble, misfire, and give up. Creating AI slop on $20 subscriptions they fill the internet with. If designers don't step into this layer, they cede the future of creativity to engineers and researchers. Which is probably why every vibe coded app looks exactly the same.
The New Math
In the new world, a model is making user-specific choices even before a prompt is written and submitted. Millions of combinations of decisions protocols that now influence every human-computer interaction. Leaving the poor designer to control little everyday. Eventually, that’s going to be zero.
Anthropic launched an incredible experiment last month that shows you software that builds itself. They’re cutting out the middle man. Not just the designer, but the engineer, and every other entity that influences the creation of software.
So let’s ask ourselves: what will become the role of a designer in a world where software is able to conjure itself?
Every technological shift creates new roles, jobs, and elevates craft.
When we look at examples like these, we need to understand one thing that’s becoming quite certain as we accelerate towards the future. The role we play today as designers might not be the same, or worse, might not even exist.
The Designer's Vanishing Point
If these inversions continue unchallenged, and right now, they are, every designer who keeps polishing surfaces is on a death march.
You can feel it already, can't you?
The slow realization that the rituals that once defined your value, the pixel-perfect mockups, the animation specs, the design system documentation, the artifacts of your glorious design title and role are starting to feel hollow. Like rearranging deck chairs. Like building sandcastles as the tide comes in. We call this busy work. Keeping you busy while intelligence takes over the digital realm and models become smarter.
Today, plenty of designers can hold intelligent conversations about AI. They follow the discourse. They experiment with Midjourney. They debate whether ChatGPT's interface is too sterile or whether Claude's voice feels more human. They sprinkle AI features into their portfolios and talk about "designing for intelligence."
But strip away the performance and ask the harder question:
How many designers are in the room where the models themselves are being shaped?
The answer is vanishingly close to zero.
The most profound technology of our time, the substrate that will mediate billions of human interactions, make decisions about health and finance and justice, is being designed without designers.
And I don't mean designed without our aesthetic input. I mean designed without our voices in the conversations that actually matter.
The New Pixels Are Invisible
When Apple built the Apple II, the original iMac, the iPhone, the Vision Pro, there was always a designer at the table making consequential decisions:
The texture of the camera plateau.
The refraction precision of sapphire glass.
The subtle choreography of the Dynamic Island's animations.
The weight distribution that makes a device feel right in your hand.
Every pixel, every interaction, every flow, and every physical curve was touched by a human eye trained in form and meaning. Designers were present at the moment of creation.
Designers imagined a world for the people/users, and designed it in the light that best represented it.
But the frontier has shifted overnight.
The design decisions that matter now don't live in design artifacts. They live deep inside the model stack. In territories most designers don't even have the vocabulary to describe, let alone the access to influence.
Let me be specific. Here are the new materials of experience design:
1. Training Data Curation
Every model is shaped by the data it learns from. What gets included? What gets excluded? What sources are weighted more heavily?
These choices hardwire worldviews into the system. If the training data over-represents certain cultures, languages, or perspectives, the model will too. If it's scraped from the internet without curation, it inherits the internet's biases—racism, sexism, misinformation, all baked into the weights.
This is a design decision with massive human consequences.
Right now, it's being made by data engineers and researchers optimizing for scale and performance. Designers, the people trained to think about representation, accessibility, and human impact, are nowhere near this conversation.
2. Context Window Architecture
How much memory does an AI have? Does it forget you after one conversation, or does it remember across months? Can it recall the preferences you stated three weeks ago, or do you have to repeat yourself every time?
This isn't a technical implementation detail. It's a fundamental design choice about the nature of the relationship between human and machine.
A short context window makes interactions feel transactional and frustrating. A long one raises profound questions about privacy and consent. Who decides where that line is drawn? Engineers optimizing for inference cost. Not designers thinking about trust and user agency.
3. System Prompts & Policy Layers
Before you ever type a word, the model has already been given invisible instructions. What researchers call "system prompts" or "constitution." These hidden rules shape the model's personality, its refusal conditions, its boundaries.
Should it act helpful but deferential, or proactive and confident?
Should it refuse certain requests outright, or gently redirect?
Should it challenge you when you're wrong, or comply even when it disagrees?
These are the wireframes of intelligence. The invisible scaffolding that determines how an AI behaves. And today, they're written entirely by researchers and policy teams. Designers, who spent decades perfecting the art of shaping tone and interaction, aren’t even thinking about these problems let alone participate in them.
4. Retrieval-Augmented Generation (RAG)
When an AI doesn't know something, it can search for information. But how it searches, what sources it trusts, how it ranks results, what it chooses to surface, shapes the entire experience.
Does it pull from Wikipedia? Academic papers? Reddit threads? Your private documents? Does it synthesize multiple sources, or just parrot the first result? These decisions determine whether an AI feels authoritative or reckless, helpful or invasive.
This is UX design at the infrastructure level. And designers aren't building it.
5. Tool Orchestration & Agentic Behavior
The most powerful AI systems can take actions: book a flight, send an email, trigger a workflow, make a purchase. But who decides the rules of engagement?
When should the AI act autonomously vs. ask for permission?
How does it handle errors or ambiguity?
What happens when two tools conflict?
How transparent should the reasoning be?
These are questions about agency, trust, and control. The deepest questions in interaction design. And they're being answered right now, in the architecture of agentic systems. Answers framed by people. Just not the traditional designers.
The Revelation
These are the new "pixels." The new materials. The invisible substrate where the future of human-computer interaction is being written.
And most designers don't even know they exist. Because AI for them is truly a black box. One they love to hate on because, “it’s never going to replace my creative spirit.” And let’s not even start with the entitled notion of, “I have better taste” or “I don’t see AI doing what I can do.”
We're still arguing about “does Figma Make augment creativity” while the actual decisions, the ones that will shape how billions of people relate to intelligence, are being made in rooms we're not invited to, using language we haven't learned, by people who will have to move mountains to help the designers of today understand these technical nuances. Which will inadvertently slow progress down. In a landscape that’s moving at blistering speed, that’s a make or break.
If designers remain absent from this layer, we don't just lose influence.
We lose our profession's purpose.
The world's most profound technology will evolve without us. And we will wake up one day to find ourselves decorating the outputs of machines that never needed us in the first place. Machines designed by people who see human experience as an optimization problem, and less so as a responsibility.
→ In Part III: Your Design Leaders Are Lying To You (dropping Oct 14)
In the third part: Why your design leaders are betraying you. The counterarguments that actually matter. And the path forward. If you're willing to take it.