→ Recap of Part II: Designers Have to Move From the Surface to the Substrate
We examined how the interface era established design as a strategic function, but also locked it into surface-level thinking. While designers mastered visual polish and user flows, the real architecture of digital experiences: algorithms, data models, system logic, remained beyond their reach. That separation seemed acceptable when interfaces were the primary medium. But as intelligence moves into the substrate, that divide has become a crisis. Design's hard-won credibility was built on frameworks that no longer map to where consequential decisions are made.
Now, we examine the institutional response, and why it's failing.
Generals of the Last War
This year's design conferences follow a reductive script. Senior design leaders, many who spent over two decades fighting to get design a seat at the table, deliver reassuring platitudes from keynote stages.
"Taste, judgment, craft will outshine everything in the age of AI."
"Our humanness is what will help us stand apart as designers."
"People will now come to realize the value of good design."
In a perpetually connected and always-online world, when Silicon Valley leaders like Altman, Tan, Amodei etc. do their usual media routes, their self-fulfilling sound bites become chants at the town square.
In design leadership circles those chants have become - Taste. Judgement. Craft. Humanness. Ethics.
To the keen observer and deep thinker, this cycle becomes predictable and hollow.
Take a step back for a moment. In the pre-ChatGPT world, were taste, judgment, craft etc. any less valuable than they are now?
No. These qualities were always isolated to a few. The Apple Human Interface team. Notion. Airbnb. Linear. Stripe. More recently, Perplexity. Quality craft isolated and led by handful of companies that acquired and empowered the best designers in the world. While the many got by imitating the few.
And no, I'm not talking about the thesis where 'because AI can now do the boring or unattainable execution tasks, these things begin to matter.' I'm referring specifically to the notion the design leaders are now projecting, almost in a self-congratulatory manner, that designers are the ones "gifted to naturally think this way."
There's a subtle shift in the narrative and tone of design leaders. Some even ego-driven. These qualities are being positioned as designers' natural competitive advantage in the AI era. The logic is seductive: because designers cultivate these capabilities, we are uniquely positioned to shape how AI systems serve human needs.
This assumes something rarely examined: that these qualities belong exclusively to designers.
They don't. Product managers with ten years of domain expertise have taste. Engineers building distributed systems have judgment about failure modes. Researchers studying model capabilities have craft in how they design experiments. What designers possessed was not exclusive access to these capabilities, but monopoly over their translation into tangible artifacts.
That monopoly is now ending.
But the big lie isn't even that the design leaders are peddling this regurgitated superfluous trope. It's that absolutely nothing is allowed to change on their teams despite a coming reckoning of a designers “busy work.”
Design leaders prescribe "use AI tools, stay curious, experiment" while keeping every ritual intact. Design sprints still run two-week cycles optimized for interface iteration. Design systems still catalog hundreds of component variations. The busy work that made design legible to skeptical organizations persists without examination, even as it grows peripheral to where actual decisions are made.
This is the lie that matters: that the existing choreography will carry the profession forward. That designers can lead in the era of AI while maintaining the workflows of the interface era.
They cannot.
If design is to genuinely leverage taste, judgment, and craft in AI-mediated experiences. If these qualities are to remain strategically relevant rather than aesthetically pleasant. Designers must go deep into an entirely new technical vocabulary. Model behavior. Training data composition. System prompts. Policy boundaries. Retrieval strategies. The intelligence layer where the new user experience is generated.
This requires more than words of encouragement. It requires institutional transformation. Leaders who use their authority to restructure sprints, redefine success metrics, reallocate time from polish work to capability building. Leaders who fight for seats in the rooms where model architectures are decided, not just where interfaces are reviewed.
And above everything else, letting the designers ship independently; while building an ecosystem for doing so. Because the tools of today enable this. Design needs to stop lurking around in onitsuka tiger shoes and matcha lattes in their hands and actually ship real ideas. Clearly, the design leaders believe we have it all. Taste, craft, judgement. So lets leverage those. Learn Cursor, learn Claude Code, and actually ship a damn thing end-to-end without the involvement of Product or Engineering. Because this is possible today. Anybody arguing it’s not, is living under a rock etched with incompetence and ignorance.
Without this, the platitudes remain exactly that: comforting words with no operational substance.
The profession is being bypassed. Not through direct confrontation. No one is declaring design obsolete. But through quiet circumvention. Strategic leverage drains to whoever controls the intelligence layer.
These leaders won the last war brilliantly. That war is over. But they're the generals of the last war. And the generals of the last war aren't necessarily going to be the ones you want leading the next one.
When Success Breeds Blindness
These leaders won the war of the last two decades brilliantly. Showing up to what was an uphill task and an almost thankless gig. They professionalized digital design between 1995 and 2020, transforming it from decoration into strategy. They built design systems, formalized research methodologies, secured executive representation. This was consequential work that fundamentally reshaped how technology companies operate.
But the paradigm has shifted beneath them, and the very frameworks that once proved their mastery now blind them to what comes next.
The structures that legitimized design, the design sprint, the quarterly review, the elaborate component library, served a specific historical purpose. They made design legible to engineering and business functions that didn't yet understand its value. They created predictability. Measurability. A vocabulary for cross-functional collaboration.
The problem? These frameworks are artifacts of an era when design's primary output was pretty much static interfaces, flows, and experiences. When designers architected roughly 85% of what appeared on a screen. When user experience emerged from deliberate choices about layout, hierarchy, interaction patterns crafted by a design team.
That assumption no longer holds.
As intelligence moves into the substrate, as models begin generating interfaces dynamically, making decisions in real time, adapting their behavior based on context designers never specified, the locus of control shifts violently. Designers in AI-mediated experiences now control perhaps less than 5% of what users encounter. The remaining 95%? Determined by model behavior. Training data. System prompts. Retrieval strategies. Policy boundaries. Inference-time decisions made in milliseconds, without asking permission.
This is not incremental change. It is a phase shift. A fundamental reordering of where design decisions are made and who makes them.
Yet the response from much of design leadership has been to double down on existing frameworks. More design sprints. More elaborate component libraries. More workshops on "AI ethics" that remain studiously divorced from the technical architecture where ethical choices are actually encoded. The rituals multiply precisely because they provide psychological comfort: the appearance of control in an environment where actual control is evaporating.
This is not a moral failing. It is an institutional one.
Leaders who spent two decades establishing design's legitimacy through predictable processes now face a landscape where those processes are increasingly peripheral. To acknowledge this, to admit that the frameworks they built, the expertise they cultivated, the organizational structures they defended may not transfer to the intelligence layer, would require intellectual humility that institutions rarely reward. It would mean relinquishing hard-won status. Starting over as students of a radically different technical substrate.
The cost of this institutional paralysis falls disproportionately on early-career designers. Those who entered the field in the past five years expecting the career trajectories their leaders promised. They are learning skills that are rapidly depreciating. Building portfolios around work that may not exist in recognizable form a decade from now. Receiving guidance from leaders who, through no fault of moral character but through structural position, cannot adequately prepare them for what comes next.
The Maginot Line
After the slaughter of World War I, France's military leadership vowed: never again. Never again the carnage of static trench warfare, the grinding attrition, the millions lost in mud. They invested enormous resources into the Maginot Line. A vast defensive fortification along the German border, embedding every lesson from the previous conflict into its design. By the standards of its era, it was a masterpiece of military engineering.
When World War II began, Hitler simply went around it.
The Maginot Line failed not because it was poorly conceived, but because the war changed. Hitler's army moved through Belgium and the Ardennes forest, a route French strategists had dismissed as impassable. Within weeks, Paris fell. The fortifications became monuments to institutional inertia, symbols of leaders so focused on refighting the last war that they never saw the next one coming.
Contemporary design leadership is building a similar line.
The fortifications here are methodological: Design sprints, component libraries, OKR alignment, demo showcase reels, ethos decks, AI design principle documents, ethic decks that demonstrate design's business impact. These structures once served a purpose. In a world when software can conjure itself, not so much.
The battle has moved. We're guarding the same old shit, while AI is moving around us.
The transformation determining design's future is not happening at the interface layer. Yet design teams continue manning these defenses. Refining their design systems. Running more sprints. Not because these activities are strategically sound, but because they are legible. They fit within existing organizational structures. They can be measured, reviewed, presented to stakeholders who understand this vocabulary.
Meanwhile, the decisions that will shape how billions of people interact with intelligence, how models behave, what they refuse, how they reason, when they act autonomously, are being made in adjacent buildings. By teams who don't speak this language. And don't feel they need to.
Design risks the same fate as the Maginot Line. Not through direct confrontation, but through circumvention. The work will continue. The rituals will persist. But the strategic importance will drain away, redirected to whoever controls the intelligence layer.
The Shallow Adoption
Observe what passes for AI integration in contemporary design practice.
Midjourney for mood boards. ChatGPT for placeholder copy. Figma Make to generate layout variations. Most teams have adopted generative tools at the periphery. Productivity enhancements, not paradigm shifts. They accelerate existing workflows without fundamentally transforming them.
This is shallow adoption. The organizational equivalent of installing solar panels on a coal plant.
Why does this pattern persist?
Deep AI integration. The kind that would genuinely transform design practice, requires designers to develop AI fluency. One that none of the respected design leaders actually have. Because they were too busy being the outside voice in a room of deeply technical minds. And you can't disseminate what you do not possess. So "learning new tools and experimenting with AI" is never going to be enough. It requires a fundamental shift in not just vocabulary, it involves a rethinking of leadership.
This superficial encouragement leads teams to experiment at the edges. Celebrate small productivity gains. "Look, designers can make real prototypes now instead of Figma noodles." Mere scaffolding while avoiding the more disruptive question: should we burn all our processes to the ground?
This is how institutions behave when faced with existential uncertainty. They perform adaptation while preserving core structures. Adopt the vocabulary of transformation while avoiding its substance.
The problem is that shallow adoption creates vulnerability disguised as progress. Teams believe they are preparing for an AI-mediated future when they are merely automating parts of a diminishing workflow. Meanwhile, product managers with strong business intuition are learning to use AI tools to bypass design entirely. Startups are building end-to-end products, proving new business ideas without ever opening Figma.
This is the bitter irony of shallow adoption: while design teams congratulate themselves for integrating AI into existing workflows, adjacent functions are using the same tools to make designers optional. So when Dylan Field talks to you about how "AI will lower the floor and raise the ceiling" you're forgetting that the room just got a whole lot bigger for anyone to use AI to do the same shit a designer would do, only much faster. Maybe not as brilliantly as a Claudio Guglieri or a Kari Saarinen, but the point here is that exquisite polish can now most certainly come after an idea has been proven. Built with their own unique taste and judgement and craft. Redefining what these things actually mean. Because the people who are actually have the strategic sense to bring a product to market will be the ones who will take advantage of this. Something most designers haven't necessarily developed as part of their skill because they were too busy perfecting layouts with autoflow and variants.
The Versailles Problem
If the Maginot Line represents defensive structures built for the wrong war, Versailles represents something more insidious: performance masking decay.
Louis XIV's court glittered with ritual. The Sun King's daily levée. His ceremonial rising was attended by dozens of nobles in elaborate costume, each with prescribed roles. Who could hand him his shirt. Who could hold the mirror. Who could light the candles. All precisely choreographed displays of order and hierarchy.
Inside the palace, everything appeared magnificent. Outside, the treasury was depleting. The peasantry faced famine. But the performance continued. Because the performance was the point. The rituals didn't merely represent power. They constituted it.
Contemporary design leadership has constructed a similar edifice.
Consider the typical design summit. Keynote presentations with cinematic production values about "the future of human-centered AI." Panel discussions where leaders in carefully curated Japanese-inspired wardrobes discuss "maintaining craft in the age of automation." Workshops on "AI ethics" that generate frameworks and principles but rarely touch the actual technical systems where ethical choices are encoded.
These are not conversations about transformation. They are rituals that demonstrate continuity. Signals to the broader organization, and to designers themselves, that the existing order persists. That the frameworks that brought success remain valid.
The disconnect becomes visible when you compare these performances to the actual conversations happening in AI research labs. There, the questions are concrete. Technical:
- How should we weight different data sources during training to avoid demographic bias?
- What refusal behaviors should be encoded in system prompts, and who decides the boundary conditions?
- When should autonomous agents act independently versus requesting permission, and how do we make that decision legible to users?
These are design questions in the most fundamental sense. They determine how intelligence behaves. What it values. Whose needs it prioritizes. They are the contemporary equivalent of choosing a typeface for a digital interface, except the consequences scale to billions of users and the decisions are largely irreversible once encoded in model weights.
And designers are almost entirely absent from these conversations.
Not because researchers are hostile to design input. Most would welcome collaboration with someone who deeply understands human needs and can translate them into technical constraints. The problem is that most designers don't speak the language of this layer. They can articulate user needs in the abstract, "the system should feel trustworthy," "users need transparency", but cannot translate these into actionable specifications at the architecture level leaving the other crafts to interpret their meanings. And this collaboration is very rare. Even though when it happens, is undoubtedly magical.
Meanwhile, the summits continue. The aesthetic of design leadership becomes more polished even as its strategic leverage diminishes.
This creates a particular kind of institutional tragedy. Early-career designers watch their leaders perform confidence and assume this represents genuine mastery. They believe that if they develop the right combination of craft skills and ethical awareness, they will navigate the transition successfully. What they cannot see, what the performance is designed to obscure, is that their leaders are equally uncertain, equally disoriented, and equally unequipped for what comes next.
The performance isn't intentionally deceptive. It is structurally necessary. Leaders who spent decades building credibility through visible excellence now face a transformation where that excellence has diminishing returns. To admit uncertainty would undermine the very authority that allows them to lead.
But principles without implementation are philosophy, not practice. And in the contest between philosophy and technical architecture, architecture wins. The systems are being built now. The defaults are being set. And when designers are excluded from these decisions, not through malice but through incomprehension of the technical substrate, those defaults will reflect the priorities of whoever is in the room: engineers optimizing for performance, researchers optimizing for capability, product managers optimizing for engagement.
User needs, if they enter at all, will do so through crude proxies and afterthought audits, not through the kind of deep, iterative design thinking that shaped the era of interface design.
Workflows Are Changing
Google, only last month, laid off hundreds of designers. It's biggest ever, where entire design teams vanished over night. Madhu Guru, a product-leader on the Google Gemini team tweeted this recently.
"At Google, we are moving from a writing‑first culture to a building‑first one. Writing was a proxy for clear thinking, optimized for scarce eng resources and long dev cycles - you had to get it right before you built. Now, when time to vibe-code prototype ≈ time to write PRD, PMs can SHOW not tell. Role profiles are blurring, creativity and building are happening in parallel." — Source (Twitter)
Meta, also laid off hundreds of designers, and continues to.
"Director in org declares designers are optional, suggests PMs use AI for product design. One of our directors recently gave their org the green light to start designing features and products without involving designers. PM told me they would consult with designers “as needed”." — Source (Blind)
These are not isolated incidents. They represent a pattern. As AI capabilities expand, companies are discovering they need fewer people in certain roles. Not because those roles lack value, but because the work can increasingly be performed by smaller teams augmented by AI tools. Or bypassed entirely through new workflows.
The most significant change is happening quietly, in the daily operations of product development. At several major technology companies, product managers with deep domain expertise have begun moving from concept to functional prototype without involving design teams in early iterations. The workflow:
A PM with years of product intuition uses Claude Code or Cursor to generate a high-fidelity prototype, describing interaction patterns and user flows in natural language. The AI produces functional code, often with components already hooked into the current design tokens, that can be immediately deployed for internal testing. The PM conducts rapid user research, iterating on feedback by prompting the AI to modify the prototype. Only after several rounds of validation does the work move to a design team, now tasked primarily with refinement and consistency rather than foundational direction.
This is not hypothetical. It is described in internal communications, corroborated by designers at these companies who have watched their role in early product development shift from central to peripheral.
I'd go as far to argue, that even this workflow is a bit of an overkill. Established design libraries already exist in most mature orgs. Nothing should ideally stop a strategic individual to context engineer a fully functioning idea and launch it after security reviews. Depending on the maturity of the org, this is very likely to happen very soon.
The implications are profound. For decades, designers held a monopoly on translating abstract product concepts into tangible artifacts. This monopoly created structural power. You could not effectively develop and test product ideas without designers to make them real.
That constraint is evaporating. And frontier leaders are already changing the workflows to fit this new construct.
Where the Capital Is Flowing
Compensation data provides a stark measure of where companies believe future value will be created.
According to levels.fyi's 2025 data, senior AI researchers and engineers at frontier labs - OpenAI, Anthropic, DeepMind, regularly command total compensation packages exceeding $600,000. Principal and staff-level roles reach $800,000 to over $1 million. In some cases, leading researchers have received offers approaching $10 million annually when equity grants are included.
Compare this to design leadership compensation. The highest-paid design executives at Meta, Google, and Apple, at VP-level roles representing the absolute apex of the profession after 20-25 years of career progression, typically reach $800,000 to $1.2 million in total compensation. Senior design directors, the level below VP, typically earn $350,000 to $550,000.
The median AI researcher at a frontier lab, mid-career, earns more than the executive design director.
Read that again.
This is not a statement about fairness or individual worth. It is a statement about market perception of future value. Companies allocate scarce capital to capabilities they believe will drive competitive advantage. The compensation gradient reveals where that belief concentrates.
How many millions did Zuckerberg drop for Design Leadership in his AI Superintelligence Lab?
→ $0.
Hiring patterns tell the same story. OpenAI, Anthropic, and DeepMind are aggressively hiring for roles in alignment research, model architecture, inference optimization, safety engineering, policy design. They are not, by and large, hiring design directors to shape how these systems interact with users.
But design leaders will be quick to quote OpenAI's highly publicized partnership with Jony Ive. This is more a signal of the kind of company OpenAI wants to build. With their announcement yesterday to build custom AI chips with Broadcom, OpenAI is simply trying to become the new Apple. So their investment in Ive's IO is more market signal, than design's organizational centrality.
In other words: PR, not strategy.
This is the uncomfortable arithmetic. Companies building the intelligence layer that will define the next computing paradigm are investing heavily in certain capabilities and largely ignoring others. They are hiring people who can shape model behavior, not people who can shape interfaces. The market is delivering a clear verdict about where leverage lives.
The Economic Logic
Technology companies are currently investing hundreds of billions of dollars into AI infrastructure. This capital allocation is predicated on a specific thesis: that controlling the intelligence layer will determine market position in the next computing era.
In this context, every non-essential cost becomes a target for optimization. And "essential" is defined narrowly: does this capability directly contribute to building, training, or deploying intelligent systems?
By that criterion, interface design, particularly surface-level interface design, begins to look expensive and slow. Not because it lacks value in isolation, but because when models can generate interfaces dynamically, when A/B testing can optimize conversion automatically, when product managers can prototype sufficiently for testing purposes, the marginal value of traditional design labor declines.
An AI-native high-agency individual can today point their Copilot to a Linear, Airbnb, Stripe website and have a clone before even finishing their morning coffee.
The hiring calculus is already shifting. Organization leaders who previously built teams of specialists, researchers, visual designers, prototypers, systems designers, will soon start asking a different question: Can this person ship independently?
Not "ship" in the traditional sense of collaborating with engineers and product managers. But ship end-to-end: zero-to-one ideas, market experiments, product validation, getting to launch without cross-functional dependencies. The designer who brings only a subset of skills that enable only a subset of collective outcomes is becoming a liability, not an asset.
These high-performing individuals are rare, but they exist. And they're disproportionately younger designers with less to unlearn. Those who naturally default to AI-native workflows rather than fighting to preserve interface-centric ones.
What about craft? What about excellence?
The emerging model is pragmatic: let AI-native individuals prove market traction with rapid experiments. Once an idea demonstrates demand, allocate capital to specialist firms: Pentagram, top-tier agencies, to bring that final 10% of polish. Craft becomes a purchased service for validated ideas, not a core competency for every team member.
This isn't theoretical preference. It's economic inevitability. Individuals who cannot leverage AI to create end-to-end products will be systematically carved out of systems optimizing for velocity and outcomes.
The result is a profession increasingly vulnerable to economic optimization, with design leadership unable to mount an effective defense because they cannot articulate value in the terms that matter to current strategic priorities. So the design leaders with hundreds of designers on their teams really have to break the current wheel in order to survive the oncoming onslaught. Something we're seeing play out in the open market currently.
The Profession at the Crossroads
The transformation is not coming. It has arrived. And design leaders are doing what the nobility of Versailles did. Performance masking.
Design, as currently constituted in most organizations, faces structural obsolescence. Not because human-centered thinking has become irrelevant, but because it has been disconnected from where consequential decisions are now made.
The profession succeeded brilliantly at a specific historical task: making software interfaces more usable, accessible, and aesthetically coherent during the era when interfaces were software's primary locus of value. For roughly thirty years, from the early 1990s through the early 2020s, this was sufficient. Design shaped what billions of people experienced every day. The work mattered because the surfaces mattered.
That era is ending.
Value is shifting to the intelligence layer. To the substrate beneath the surface. And design, as an institutional function, has largely failed to follow.
The Choice
Every designer now faces a choice.
You can continue seeking guidance from leaders who are excellent generals of yesterday's war. Continue perfecting skills optimized for a diminishing paradigm, hoping the transformation will somehow spare you. That your craft will be recognized. That your leaders will guide you through.
Or recognize that the paradigm has already shifted. That the work determining design's future is happening now, in rooms where most designers are not present. And that if we want to shape that future, we must be willing to fundamentally transform what it means to practice design.
The design leaders of the past deserve respect, especially those with the self-awareness to admit they're unprepared to lead in this new world order. But commanding that respect means something different now. It means making way for what comes next. Not through words, but through action. Not through keynotes about the future, but by vigorously supporting those who are building it.
We need high-agency designers in every organization to step up and demonstrate what's possible. Waiting for design leadership to steer has proven futile. The path forward will not be charted by those who built the last paradigm. It will be shown by the designer who just walked through the door, wondering where we go from here.
The transformation is here.
The profession is not prepared.
But individual designers still have agency.
What you do with it will determine not just your career, but the future of design itself.
→ In Part IV: How Future Designers Will Win With AI (dropping Oct 21)
In the last part: we examine what transformation actually looks like. Not platitudes about "staying curious" or "experiment with AI," but concrete actions for how designers can build genuine capability at the intelligence layer. How to acquire fluency in model behavior. How to ship end-to-end without waiting for permission. And why the path forward requires abandoning much of what made design successful in the interface era.