→ Recap of the Series so far
Parts I-III established the crisis. The paradigm shifted. Designers control 5% where they once controlled 85%. Design Leadership is paralyzed. But critique without action is performance. This final part outlines three paths to sovereignty: Brand, Product, and Intelligence. And what it takes to seize them before the window closes.
A Return to Sovereignty
In 1450, a blacksmith in Florence possessed something that very few knowledge workers in our modern society possess: sovereignty. From raw iron to finished horseshoe, from commission to delivery, they controlled the entire value chain. They didn't need permission from a VP of Horseshoe Strategy. They didn't wait for the Metallurgy Team to approve their approach. They didn't coordinate through six layers of management to validate market fit. They had iron, a forge, a hammer, and skill. That was enough.
Every craftsperson, from a carpenter, a weaver, a glassmaker all operated the same way. They took raw materials combined it with human need and produced finished value. Their work was complete and finite. Their contribution, direct. Their relevance, atomic, but undeniable. For most of human history, individuals who possessed complete capability stacks could create value without asking for permission.
The Industrial Revolution shattered sovereignty. The Information Revolution, made it exponentially worse.
Work that one person could do became work that required coordination across dozens, then hundreds, then thousands of people. A single product like a car, a smartphone, an app, now requires so many specialized skills that no single individual could possess them all. You need engineers who understood metallurgy but not electronics. Marketers who understood customers but not manufacturing. Accountants who understood finance but not product development. Sovereignty fragments into dependencies within the modern enterprise.
And when work fragments into dependencies, you need something to coordinate them. This resulted in the emergence of the Org Chart in modern enterprises. Boxes and arrows showing who reports to whom, who has authority over what, how information flows upward through hierarchies. For 150 years, this was the architecture of work. It was never meant to be permanent. It was a response to human limitations. Our inability to possess complete capability stacks, and our need for coordination when work exceeded what any individual could do alone.
This fragmentation is why Design within modern organizations today is a word salad of different specializations ranging from Experience Design, Visual Design, Product Design, Interaction Design, Motion Design, Service Design and the list goes on. Product Management splits across Business and Technical, while Engineering again has a long list of specializations.
But maybe now with the Intelligence Revolution, we could return to our sovereign roots. As coordination costs go down to zero, we could technically delegate the bulk of our work to an AI system, while choosing to redefine the value we as individuals have in society. An environment where roles and titles take a back seat but the outcomes an individual creates beings to shine. And maybe, just maybe, Designers around the world currently polishing surfaces, fighting the system to take a more human lens, asking for permission, can begin to take more ownership of the direction their work takes. Shaping the world in the light they have been trained to see the clearest.
A return to sovereignty, especially for designers is now more than ever, an imperative if we're going to win in the age of AI. One that might be equal amounts liberating and fulfilling.
What Sovereignty Means
sovereignty (noun)
supreme power or authority
For an individual, the concept of sovereignty simply means having the right and moral authority to control your own life, actions, and body. Free from external control by others or by the state.
Sovereignty rests on three capabilities that AI is now making possible again.
Ownership
Control over your output and decisions. Imagine being able to decide what to build, when to build it, how to approach it, without waiting for permission from product managers, alignment from stakeholders, or approval from six layers of review. Your work becomes yours again. Not in theory, but in practice. AI gives you the capability stack to validate your own ideas, ship your own features, prove your own hypotheses. You answer to the market and your craft, not to coordination overhead.
Autonomy
The ability to act on your judgment without constant permission or coordination. When you see an opportunity or identify a problem, you don't need to schedule a planning meeting, build a business case, or wait for the engineering team to have capacity. You assess the need, make the decision, and execute. The authority to act lives in you, the person doing the work. AI collapses the gap between intention and execution. What once required a team now requires vision and orchestration.
Responsibility
When you control the entire value chain, you own the outcomes. There's no one to blame for poor requirements, technical debt, or misaligned priorities. If what you build fails, it's your judgment on the line. This creates tight feedback loops between decisions and consequences. You learn faster. You improve faster. Your craft becomes accountable in ways that fragmented work never allows.
Together, these three capabilities create something powerful: the ability to create value without permission, and taking responsibility for your outcomes.
The Three Types of Sovereignty in Design
In an AI-native world, there are three archetypes designers can embody. Before we introduce them, the big thing to caveat is stating that in no way are these three meant to be “job titles.” That would defeat the purpose and we will fall back into the trap of roles/definitions/titles from the world the org chart created.
The purpose of shaping these three types of sovereignty is to provide a framework and stepping stone for how the people who have design titles today could potentially see themselves taking one or more of these three paths. A lot of people like to return to the comfort of what they know. That's simple human nature. But the nature of work is undergoing a paradigm reinvention. So while we certainly can draw upon what we know, we can't and should not use it to inform the future. Because doing so will simply slap lipstick on a pig. Rather, we need to create something entirely new.
1. Brand Sovereignty
A good brand, a well designed brand, a more mindful brand, a caring brand, an honest brand isn't going anywhere. Apple in many ways has elevated customer expectations to a level where the customer now expects a level of polish and craft from almost every touch point they experience. Whether that is an app, a physical store, or even an offline service, customers have been conditioned to expect more before they part with their wallet.
But all along, the way beautiful and exquisite brands have been envisioned and created belonged in shops like Pentagram (including the smaller niche shops) that took advantage of a reputation they've built over the years by carefully curating the best designers in the world with a strong reputation for extremely high quality of work. From Paula Scher to Georgia Lupi, the density of talent and the output is undeniable in firms like these. This is why companies pay millions of dollars to elevate what their brands stand for.
Over the years these firms have developed a process and craft unique to them that lets them take a vision to heights very few people can actually achieve. But they're restricted by one main thing constantly. Scale. Exceptional designers cannot scale themselves. They simply cannot spend hours on projects that already have tight deadlines and budget constraints to create the vision they want to. They have to rely on other designers who adopt their vision, after several excruciating internal and client reviews directions get approved. Post which they have to rely on an army of designers and other disciplines to execute the vision. From several mockups, to renders, to treatments, to environmental impressions, they need design peers and young designers for support. Young designers learn through this process of participating in building a brand language. Working on their attention to detail, understanding the process, the ropes, and eventually proving themselves enough to being offered to lead a creative project some day.
The level of coordination, team size, back and forth in this construct of work is something AI is meant to disrupt. But not just coordination, the art itself is now something that's entirely up for interpretation. Lets break this down.
What was the one thing holding behind an exceptional designer? Scale. AI is exceptionally good at helping people scale themselves. So someone like a Georgia Lupi can work closely with an AI model she's customized to her tastes to achieve that scale. All she has to do is point the AI model in a specific direction to help visualize a plethora of different outcomes or treatments. She can design a single logo once, and see different creative variations of that same logo. Not just variations, but a visual model could potentially leverage several different assets from the Pentagram libraries to visualize that logo in another plethora of environments. Something that would take a team of 8-12 designers to envision over 4-6 weeks is now a few minutes away. As an exceptional information designer Lupi has a keen eye for how data can be made beautiful. The AI system could leverage her existing work to visualize data from several public data libraries in scenarios Lupi would not have even imagined; helping her develop a foresight into her own work she knew not she could explore.
So what happens in a world where exceptional designers like Lupi are going to rely on an AI system rather than a team of starry eyed young designers? How will the young designers learn if the team constructs are disrupted by AI?
Frankly, they do not need to worry. An org chart creates these ridiculous artificial assumptions that someone has to learn and come up the hard way. Do the work, deep in the trenches before they can own something and have primary attribution. These structures not only put designers down, they also curtail ambition. Armed with AI, the next Georgia Lupi might be discovered while she's still discovering her own craft.
A young designer studying at Parsons, or even in school today, maybe even in a different country not going to school, could collaborate with an AI system, similar to the one Lupi uses, and do the same envisioning Lupi does but in a context they best see fit. While in the old model things like craft, observation, and context curation was something that would take time to develop, in the new model an AI could partner with you to help you see your work in better light. It is very much possible today for a young designer to leverage an AI model, point it to Lupi's work, or even Pentagrams entire archive and create something new with a lens unique to them.
We've seen this more visibly and successfully achieved in the creator economy, where creators like Life of Riza (Kariza Santos) started at a very young age, are documenting their journey and have achieved millions in following. Riza isn't a world renowned filmmaker, but has a distinctive style in storytelling, lighting, and cinematography that sets her apart from even the best. The same will be true for the young designers who become AI native first. The opportunities they will have will be far more incredible that someone like a Georgia Lupi, who came up the hard way in the world of design would only dream to have.
That's not to say things like copyrights and trademarks aren't important. They absolutely are. You could very well apply the "good artists copy, great artists steal" metaphor here. But the power to replicate work at scale and in an instant is absolutely possible today. You can do this today with the technology even in its current rudimentary state. You like the 3D style Airbnb icons or their 3D style video ads? Drop all references from Airbnb or Buck into a model like V7, Aleph, or Veo 3, and voila you have your own custom airbnb-styled brand assets. The quality of these models has leapfrogged exponentially and will continue to get better in the next few years.
Some may argue that this model-led creativity will continue to commoditize what designers do. Please wake up. Creativity has been on the commoditized trajectory for a long time. Squarespace, Wix, Canva, Midjourney, Veo 3, and will continue to bite off larger chunks of the creative process. We simply have to find a way to adapt in this new world order while continuing to adapt not just our tools, but even our processes. Emmette Shine has an incredible breakdown of how he's reshaping the work he's always been doing to adhere to this new world order.
But more importantly, and this is the biggest point of this narrative shift we're experiencing, is the inherent sovereignty this will enable for the designers who continue to pursue Brand as a discipline in their careers. While Brand Design isn't going anywhere and will continue to become even more important in a world where everything begins to morph into a sea of sameness, the sheer opportunity individuals will have to create, direct, and distribute their work impacting the next generation of brands will be significant; despite the commoditization of creativity. While this type of sovereignty may not be for everyone and not very well compensated, this is undoubtedly one of the more resilient archetypes that will continue to exist. This will fundamentally shift the narrative from firms like Pentagram becoming the epicenter of exceptional brand design to individuals like Lupi and Shine, who brands will want to directly engage with because AI has augmented them in multiple folds; something that wasn't possible before the era of intelligence.
2. Product Sovereignty
One could argue every app, software, product, or service that needs to exist, already does. The app store and the web have had decades to evolve and you can find even the most niche services now influenced by software. Software has “eaten” every industry and every type of business. From CRM to Design tools, Enterprise Search to even Plant care, we have very high quality software across the board. Interfaces have been perfected, workflows have been optimized, and market leaders have been crowned. For thirty odd years, software innovation meant building better versions of things that already existed. Faster load times, cleaner interfaces, more features. Incremental improvement over incumbent products.
That era is now over.
Artificial Intelligence is rendering entire categories of software obsolete under the new paradigm. Forcing us to reimagine software at a very fundamental level through first principles. As companies race to bolt AI into their apps, the relentless speed is forcing designers today to resort to its comfort zone by reapplying patterns from the old model. Notion slaps an assistant to the right, albeit with cute animated characters. An incapable Siri struggles to keep up with ChatGPT while leaning on a colorful ethereal glow. This is surface-level thinking. The trap that makes every AI product look the same today. Mere scaffolding that wraps the archaic model in new drapes.
The thesis in Silicon Valley is we’re still in the very early days of AI transformation. And because most “early apps start out like toys,” that’s what we’re currently seeing. The first wave of generative AI applications “resemble the mobile application landscape when the iPhone first came out; somewhat gimmicky and thin." Where instead of leveraging the fundamental capabilities of smartphone hardware, developers just wrapped websites in apps. That said, both Apple and early developers like Flipboard nailed the model that set the standard for what apps would eventually become. Rich, native features that elevated experiences beyond the web. We are yet to see innovation in the AI landscape that truly reshapes the category of what an AI-native application should look like. We're still focused on making existing software and tools AI-capable. E.g. AI (bloody) browsers - ChatGPT Atlas, Perplexity Coment are focused on AI-enabled browsers rather than asking the fundamental question of “why do we need browsers anymore?”
Real reinvention means asking: if this category were invented today, knowing what AI can do, would it look anything like the incumbent?
Would "legal research" still mean searching databases of case law, or would it mean an AI that reads contracts, identifies risks, and drafts amendments autonomously? Would "customer support" still mean ticket queues and response times, or would it mean an agent that resolves 80% of issues before a human ever sees them? Would "project management" still mean Gantt charts and standups, or would it mean an AI that attends meetings, updates tasks, and generates status reports without being asked? Would the "word processor" still look like Microsoft Word if it were invented using LLMs today?
While every single startup founder and enterprise software executive is busy racing to compete with each other, ship aggressively, this leaves the door wide open for every single product designer on the planet to ask the question, that frankly every one should be asking. What do AI-native applications look like? Because they sure as hell don't look like the ones we're all racing to build. And above all, this is an opportunity that comes with, for the very first time in history, for all talented product designers to not just envision, but build the entire reimagined experience and launch it. Serendipitously creating the opportunity for sovereignty; for those who are brave enough to reach out for it. It is now, more than ever in the history of software easy to raise money and build a startup quickly and be recognized for it. And product designers, the talented people who have shaped world class tools should be all over this opportunity. Alas! all they have to do is get their heads out of Figma because nothing real can be shipped through it.
The market knows this and is begging to see first principles led innovation. Ycombinator has been shouting this from the rooftops for over two years now. "We've been too focused on model innovation. Where is the AI application stack?", "Vertical AI Agents Could Be 10X Bigger Than SaaS." The message is unambiguous: Everyone has been so focused on model innovation that the real opportunity has shifted. It's moved up the stack. The application layer is wide open. The window seems to be closing as OpenAI, Anthropic and others have realized model parity is not leading to any more differentiation. So the most logical approach now would be to go broad across the application landscape by leveraging their brand, and go deep into vertical strategy like Health, Finance etc. The differentiation isn't happening at the model layer anymore. It's happening in how you apply intelligence to specific domains. Entire industries with workflows that haven't fundamentally changed in decades. AI isn't making those workflows 10% more efficient. It's replacing them entirely.
Harvey AI is currently valued at approximately $5 billion, following its Series E funding round of $300 million. Founded by Gabriel Pereyra, who studied computer science, had zero legal training or ever stepped into a law firm. Harvey today is the flagship AI example of legal intelligence. It doesn't just augment legal research, it replaces the archaic workflows that have defined law firms for decades.
The paradigm shift: Law firms operated on a pyramidal labor model where junior associates spent 60-70% of their time on pattern-matching work: contract review, legal research, due diligence, document drafting. Harvey doesn't just make this work faster. It performs it autonomously and delivers finished output for partner review. What required teams of associates billing hundreds of hours now requires one senior lawyer verifying Harvey's analysis in a fraction of the time. The AI-native solution isn't "better search" or "assisted drafting." It's eliminating the need for human labor on the repetitive cognitive work entirely, collapsing the billable hour model that has sustained law firms for a century.
As of October 2025, Harvey has 74,000+ lawyers using the platform across 700 law firms and enterprises, including 50 of the AmLaw 100 firms. Major clients include PwC UK, Deutsche Telekom, A&O Shearman, and Repsol. They're buying it not because it makes lawyers more efficient, but because it replaces entire tiers of legal work. The product is the workflow replacement. The interface is secondary to the intelligence.
What's happening here isn't the simple fact that it is now exponentially easy to use AI to replace all the tedious, mundane, and soul crushing work of the associates. It's the approach Harvey has taken to create a vertical solution that simply cannot be replaced by the model providers. Creating a category differentiation factor by eliminating the cronos effect where each new release by OpenAI or Anthropic could eat up their business. Harvey's solution represents what fundamentally is a carefully chosen design problem where the application of technology is not just sublime, but also a monopolistic and profitable one. And for those who argue this leads to hundreds, potentially thousands of associates who will find themselves out of a job. Why don’t you take a moment to ask the same associates how grueling their jobs was prior to Harvey. This means these same associates now can focus more on strategy than being glorified google search experts.
Problems like the ones Harvey solves is right up the alley a trained designer eye is meant to see in the world. With enough understanding of the technology baseline, a good designer must be able to find opportunities out in the wild to completely disrupt entire industries. When someone talks about “design being a differentiator” and “how design led companies are the future”, we keep resorting to the tired examples of companies like Airbnb, Notion, and Linear. While only one of them is part of the Fortune 500. Designers seem to be rather content with these three outliers amidst a universe of stellar companies.
Designers need to understand that this is the opportunity of a lifetime to achieve Product Sovereignty. We need to understand that Product Sovereignty diverges completely from the old paradigm. It's not about designers learning to code so they can communicate with engineers better. Designers can now delegate all the code to a model. It's about designers leveraging what they do best to reimagine entire constructs; building companies that replace entire categories of work with intelligence. Not tools. Not features. Complete workflow replacements. And fundamentally because everybody is racing to compete in simply being seen with an AI copilot bolted onto their archaic software, none of the people actually making new stuff are pausing to reassess. Designers have probably another year before the market realizes this at large. We’re already seeing radically new products emerge. Designers need to find a way to rapidly reskill, away from pixel pushing surface tools, and dive deep into partnering with AI on outcomes-led thinking that helps them ship real software; evolving their capability stack.
The capability stack required to do this is now accessible to individuals in ways that were impossible a mere two years ago. Cursor/VS Code, Claude/Sculptor let you build full-stack applications without knowing how to code. Replit Agent handles deployment, hosting, database setup. Vercel and Railway abstract away DevOps. Stripe handles payments. The entire infrastructure layer that used to require a team of engineers is now “orchestratable” by one person who understands what needs to be built and why. These technologies are a gift to designers, if only they realize.
But, and this is critical, building is not the hard part anymore. AI can generate all the code one needs, and you don’t even need the best most powerful model anymore. AI can debug. It can deploy. What it can't do is identify which workflows are ripe for replacement. It can't understand the institutional friction, the coordination overhead, the manual processes that professionals tolerate because "that's just how it works." It can't see the category that doesn't exist yet. That's the designer's advantage. Designers are trained to observe how people work, where systems break down, what needs aren't being met. For 20 years, that skill was channeled into making interfaces prettier or coordination smoother. Now it can be channeled into something far more consequential: identifying the next vertical AI category and building the company that owns it.
Ycombinator is running an investment thesis of AI native companies asking the question if "10 People + AI = Billion Dollar Company?" People are raising capital easily with a solid understanding of industry-led vertical workflows. Most founders of these new companies won’t come from Big Tech or AI Labs. They need to emerge from the companies designers are already helping shape in the age of AI. The people who lived the workflow, understood its inefficiencies, and saw what intelligence could replace. Designers already embedded in workflows have had full visibility of the internal systems and working functions of large companies. Walmart, Target, Nike, Mercedes Benz, Home Depot, KKR, Goldman Sachs. These designers have watched engineers, PMs, marketers, salespeople, support teams struggle to ship features for years because of the complexity of businesses. They already understand the inefficiencies of these ecosystems. The coordination overhead, manual processes, fragmentation of tools etc. You know which workflows are ripe for replacement. The only question is whether you're willing to stop designing for companies and start building them yourself instead.
Product Sovereignty in this context isn't about shipping features autonomously. It's about identifying category-defining opportunities and building the companies that redefine problems. It's about seeing that legal work is pattern matching, that enterprise search is a retrieval problem, that project management is orchestration. And having the capability stack to turn that insight into a working product, a real company, a venture-scale outcome. The window is open right now. Foundation models are good enough. The tooling is accessible enough. The market is ready. But it won't stay open forever. In 12-18 months, the obvious vertical AI categories will begin to become saturated.
The designers who achieve Product Sovereignty won't be the ones who learned to code so they could avoid waiting for engineers. They'll be the ones who saw a category that didn't exist, built the AI application that created it, and became the founder. Not a feature owner. Not a product lead. A company builder. That's sovereignty in the age of intelligence.
AI is already slowly creating a significant wealth gap in the industry and the economics of the world. We see news about large payouts to AI reseachers and engineers. In the product-design-engineering trifecta designers are already valued and comped lower down the totem pole. If designers don’t skill up, build radically new constructs, that are vertically positioned we might not have another opportunity to achieve sovereignty again at this scale. Designers need to shift their mindset from owning artifacts to creating real economic value. Those who do will have something the traditional product designer never imagined possible: ownership of a category, a company, and a future they built entirely on their own terms.
3. Intelligence Sovereignty
The final kind of sovereignty is one of Intelligence. Something designers have been explicitly held back from as they’ve traditionally polished surfaces. These are the rooms designers haven’t historically entered (summarized in parts I-III). The hardest part of this transformation isn't the technical learning curve. It's the category confusion. We don't yet have a shared language for what substrate design is. In order to make this easy, define what has been undefined for the last 2-3 years, let’s break this down through a set of examples related to where designers can lead in AI-native companies. Specifically focused on how designers can reshape how billions of users interact with intelligence.
Example 1: Anthropic’s Constitutional AI
Designing Personality at Scale
At Anthropic, a small team shapes Claude's personality not through interface polish, but through constitutional AI. A system where the model's behavior is governed by principles encoded directly into its training process.
This isn't copywriting. It's architecture-level design: deciding which human values the model should optimize for, how it should handle conflicts between competing principles, when it should refuse vs. redirect, how transparent its reasoning should be. The artifact isn't a mockup. It's a constitution. A structured set of rules that shapes every response Claude generates across billions of interactions.
Someone had to decide: Should Claude be deferential or confident? When a user asks for harmful content, should it refuse coldly or explain warmly? How much uncertainty should it express when it's not sure? Should it optimize for helpfulness or for protecting users from themselves?
These are the substrate equivalent of choosing button placement and color palettes. Except they scale to billions of users and can't be A/B tested away if you get them wrong.
Where Designers Fit: Constitutional AI is designed by alignment researchers. People with PhDs in ML who understand model training at a technical level designers simply don't possess. But there's a gap. Researchers know how to encode principles into training loops. They don't necessarily know which principles matter most to users, how conflicts between principles feel in practice, or how to communicate constitutional boundaries in ways humans understand. This is why you end up with glossy sentences like “As an AI language model…” or “You’re absolutely right.” A designer who can understand sufficient ML to speak the research language can become extremely valuable. By helping researchers translate human values into clear testable specifications.
Example 2: Perplexity's Citation Architecture
Designing Trust Through Retrieval
Perplexity's defining feature isn't its interface. It's how it retrieves and cites sources making the search functionality unparalleled across the industry. Every answer is grounded in real-time search, with inline citations that let users verify claims instantly.
This seems simple. It's not. Someone had to design: Which sources to prioritize (academic papers vs. news vs. Reddit?)? How to rank conflicting information? When to show uncertainty vs. confidence? How to make citations feel informative without overwhelming? What happens when sources contradict each other?
The artifact is not a UI, but a retrieval policy. The logic that determines which information surfaces, when, and how it's weighted against other sources. In a world where hallucination is the default failure mode, Perplexity made a substrate-level design choice: ground every answer in verifiable sources. That decision shaped the entire product in ways no interface redesign ever could.
Where Designers Fit: The RAG pipeline, retrieval, ranking, citation, is all built by ML engineers who understand vector embeddings, semantic search, and ranking algorithms. Designers don't build this. But someone has to decide: when should we show 3 sources vs. 10? How do we present conflicting information without paralyzing users? What does "confidence" mean to someone who isn't a statistician? These are questions engineers can answer technically but often solve poorly for humans. A designer who understands RAG architecture enough to ask informed questions can help shape these decisions. Not by owning the pipeline, but by being the voice insisting that "technically correct" isn't the same as "useful." You're not the ML engineer. You're the person making sure their brilliant technical work doesn't ship with a garbage human experience.
Example 3: OpenAI's Refusal Behaviors
Designing the Boundaries of Intelligence
When ChatGPT refuses a request, that refusal is the product of policy design. A set of rules encoded in system prompts and RLHF that governs what the model should not do.
Someone had to decide: Should it refuse to help with homework, or just warn about academic integrity? Should it generate violent content for fiction writers, or draw a hard line? When users try to jailbreak it, should it be playful, stern, or simply silent? How many refusal categories exist, and where are the boundaries?
The artifact is a refusal taxonomy. A structured map of what's forbidden, what's cautioned, what's allowed, and the reasoning encoded into each boundary. These aren't legal compliance checkboxes. They're decisions about agency, paternalism, trust, and power. Every refusal shapes the user's mental model of what the system is and who it serves.
Where Designers Fit: Refusal systems are built by trust & safety teams, policy researchers, and RLHF specialists who understand how to encode ethical boundaries into model behavior. Designers don't own this process. Lawyers, ethicists, and researchers do. But there's a human experience problem. A refusal can feel patronizing, cold, infantilizing, or respectful depending on how it's delivered. Imagine being refused help by a model where a user has a chronic health condition struggling to cope with adherence because of Health laws. Someone has to advocate for "how does this feel?" when the room is full of people optimizing for "does this prevent harm?" A designer who understands prompt engineering and RLHF can contribute by prototyping different refusal tones, testing how users react, and helping the policy team see trade-offs they might miss. You're not building the moral operating system. You're making sure it doesn't make users feel like children when it says no.
Example 4: Replit Agent
Designing Agentic Orchestration
Replit Agent doesn't just generate code. It acts: reads documentation, runs tests, debugs errors, deploys projects. Someone had to design the orchestration logic. When the agent acts autonomously vs. asks permission, how it handles failure, how transparent its reasoning should be.
The decisions: Should it show every intermediate step, or just the final result? When it encounters an error, should it retry silently or explain what went wrong? How much context should it retain across sessions? When should it give up and ask for human help?
The artifact is an orchestration policy. The invisible choreography that determines how an autonomous agent behaves in the wild. This is the future of all software. Agents that act on your behalf. And if designers don't shape how they act. When they ask permission, how they explain themselves, what values they optimize for. The engineers building these systems will; which is exactly what is happening and you very visibly see the frustration of the users using these systems.
Where Designers Fit: Orchestration logic is code. State machines, decision trees, error handling. Written by engineers who understand distributed systems and agent architecture. Designers don't write this. But someone has to decide: does the agent show every step or hide complexity? When it fails, does it explain why or just retry? These are trade-offs engineers can implement but often don't think deeply about from a human perspective. A designer who understands how agents work, what "tool calling" means, how context gets passed between steps, can shape these decisions by prototyping different interaction patterns and pressure-testing them with users. You're not the systems architect. You're the person making sure "autonomous" doesn't mean "black box." Which is exactly what interactions with an AI assistant right now feel like.
Example 5: NVIDIA
Designing the Developer Experience of Intelligence
NVIDIA doesn't sell AI to consumers. They sell the infrastructure that makes AI possible. GPUs, CUDA, development platforms, inference engines. Their customers are ML engineers, researchers, and developers building AI systems. When NVIDIA designs, they're not designing for end users. They're designing how the people who build AI actually work.
This is invisible to most people. But if you've ever used ChatGPT, Claude, Midjourney, or any AI product, it's running on NVIDIA infrastructure. The decisions NVIDIA makes about their developer tools, APIs, documentation, and workflows directly shape what's possible for every AI company in the world.
Someone at NVIDIA had to design: How should error messages explain GPU memory failures to developers who aren't hardware experts? What should the deployment workflow look like when a researcher wants to move from local testing to production inference? How do we make model optimization (quantization, pruning, distillation) understandable to engineers who've never done it before? When should the system abstract complexity vs. expose low-level control?
The artifact isn't an interface for consumers. It's a developer experience. Documentation, APIs, tooling, error handling, deployment workflows. An ecosystem that determines whether building AI is painful or productive.
Where Designers Fit: NVIDIA's dev tools are built by systems engineers who understand GPU architecture, CUDA programming, and distributed computing at a level most designers will never reach. But there's a translation problem. Engineers build tools that work. They don't always build tools that are learnable, debuggable, or forgiving when things go wrong. A designer who learns enough about ML infrastructure to understand what developers actually struggle with. Things like memory constraints, deployment complexity, model optimization trade-offs, can shape these tools by prototyping better error states, testing documentation with real users, designing onboarding flows that don't assume expert knowledge. You're not the CUDA engineer. You're the person making sure that the most powerful AI infrastructure in the world doesn't have a learning curve so steep that only PhD researchers can use it. When every AI company depends on your tools, making them 10% more usable has civilizational impact.
What Intelligence Sovereignty Actually Means
Intelligence Sovereignty is earning the right to shape how AI behaves by becoming technically credible enough to contribute alongside experts. Something very few designers are actually doing today.
Not replacing ML engineers. Not owning alignment research. But learning enough. RLHF, retrieval, context windows, policy design. To be genuinely useful in the rooms where substrate decisions are made. Building prototypes with real code that demonstrate you understand how intelligence works. Making yourself indispensable to problems requiring both technical depth and human insight.
The designers achieving this today stopped waiting for permission to learn hard things. And they're being compensated accordingly. Upwards of $500-800k+ and sadly, there are very few working in this capacity in the frontier AI labs. Roles where a designers human-focused lens meets model behavior, where human experience shapes constitutional AI, where the artifact isn't a mockup but a policy that scales to billions. There are no Dribbble likes or fancy Framer portfolios that come with this work. No design awwwards or site inspire mentions. The recognition comes from researchers saying "your perspective made this better," which shows in telemetry of the end product. From shipping systems that billions use, and from compensation packages that finally reflect design's strategic value when applied where it actually matters.
That's sovereignty. The credibility to contribute where decisions are made, the capability to ship what matters, and the compensation that comes from being irreplaceable at the intersection of intelligence and humanity.
The End of One Era & The Beginning of Another
We began this series in the world Jony Ive built. A world where designers shaped curves and chamfers, where interfaces were the primary medium of human-computer interaction, where craft meant pixel-perfect execution and attention to invisible details that made technology feel magical.
That world is not returning.
We are now in the world people like Demis Hassabis will create. Where intelligence is the medium. Where the consequential design decisions happen beneath the surface, in model architectures and training data and policy boundaries. Where the curve of an aluminum edge matters less than the curve of a learning rate. Where the designers who succeed will not be the ones who perfect surfaces, but the ones who shape how machines think.
This transition is not gentle. Parts II and III documented the crisis: the five paradigm inversions, the institutional paralysis, the design leaders fighting yesterday's war while the profession loses strategic relevance. The evidence is not ambiguous. 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.
But crisis also creates opportunity.
The three paths of sovereignty outlined in Part IV - Brand, Product, and Intelligence are not just survival strategies. They are invitations to something more consequential than what came before. Individual designers now have capabilities that entire studios couldn't access a decade ago. The ability to create world-class brands independently. The ability to build and ship entire companies without permission. The ability to shape how billions interact with intelligence by earning credibility at the substrate level.
This is the return to sovereignty promised at the beginning of this part. Not the fragmented, permission-seeking, coordination-heavy design work of the org chart era, but complete ownership of outcomes. Like the blacksmith in 1450 Florence who controlled the entire value chain from raw iron to finished horseshoe, designers today can control the entire arc from insight to shipped product. If they're willing to learn hard things and build new capability stacks.
But time is the constraint.
The systems being built now, the model architectures, the default behaviors, the policy boundaries, will shape how intelligence operates for decades. These are foundational choices that become increasingly difficult to change as they scale. And right now, most of these choices are being made without meaningful design input. Not because researchers and engineers are hostile to designers, but because designers haven't made themselves credible at this layer.
The window is open. But it won't stay open forever.
In 12-18 months, the obvious vertical AI categories will be saturated. The foundational model behaviors will be locked in. The companies that will define the next computing paradigm will have been founded. The designers who waited for permission, who stayed in their comfort zones, who hoped their leaders would guide them through—they will find themselves on the wrong side of an irreversible transformation.
What Comes Next
Writing this series has been equal parts foresight and learning. Working at a frontier lab gives me visibility into decisions most designers never see. The substrate conversations, the policy debates, the architecture choices that shape how billions interact with intelligence. But it also forces me to confront hard questions I didn't have answers to.
Where is the signal amidst the noise? Do only designers have the advantage of sovereignty or does anybody with a human lens? What does the day-to-day look like for a designer chasing this kind of sovereignty? What does a portfolio now look like? What does AI-native design practice actually look like when intelligence is the primary medium from the start? We still have a long way to go before we begin to answer these questions.
This series diagnosed the crisis. But the prescription, the tactical path forward, requires continuous investigation. So follow along at future(memo) and subscribe. The era of surfaces is over. The era of intelligence has only just begun. Let me help you make sense of it.