Discover how AI is reshaping design careers. Learn why traditional design processes are obsolete and what skills designers need to thrive in an AI-first world.
Design Process Dead: How AI Is Transforming Designer Roles in 2025
The traditional design process that shaped an entire generation of designers is fundamentally changing. What was once gospel—extensive research, elaborate mockups, detailed prototyping—is becoming increasingly obsolete in an AI-accelerated world. This isn't a gradual shift; it's a forced evolution driven by engineering teams that can now execute ideas in hours rather than weeks.
Jenny Wen, former Head of Design at Claude and Director of Design at Figma, offers unprecedented insight into this transformation. She's experiencing the future of design in real-time at Anthropic, where product cycles have compressed from years to months, and designers must fundamentally rethink their role.
Core Insights: The New Design Reality
- The 60/40 Shift: Design work has flipped from 60-70% mocking/prototyping to just 30-40%, with the remaining time split between implementation and strategic direction
- Vision Timelines Collapsed: Design visions have shrunk from 2-5 year roadmaps to 3-6 month horizons, focusing on functional prototypes rather than beautiful decks
- Two Types of Design Work Emerge: Implementation support (helping engineers execute quickly) and strategic direction (creating vision without overspecifying solutions)
- Speed Builds Trust: Launching early features with caveats, then iterating based on real user feedback, actually strengthens brand trust more than perfectionism
- Technical Skills Matter More: Designers now need coding familiarity, not to become engineers, but to work collaboratively in an AI-native environment
The Death of Traditional Design Process
The conventional design methodology—discover, diverge, converge, refine—was already showing cracks before AI arrived. But with Claude Code, Opus models, and AI agents that can prototype features in minutes, the old approach doesn't just feel slow; it feels like obstruction.
"As a designer, you no longer have the time to create elaborate mocks," Wen explains. "A significant part of the design role now involves helping engineers and teams execute, rather than just delivering a final design." This isn't a limitation; it's a fundamental shift in what designers actually do.
The pressure doesn't come from management mandates demanding designers "evolve." Instead, it emerges from engineering velocity. When engineers can spin up prototypes by having a conversation with an AI, waiting for design mocks becomes friction. Designers who recognize this and adapt become force multipliers. Those who cling to the old process become bottlenecks.
What's particularly interesting is that this change was already inevitable before AI. Companies were already questioning whether multi-month design processes produced better outcomes than rapid iteration with real users. AI simply accelerated what was already becoming visible—that execution and learning beats perfection and prediction.
How Design Work Is Being Stratified
The future of design isn't singular; it's bifurcating into two distinct categories with very different skill requirements.
Implementation and Execution Support: This is where most design work happens now. Engineers are building features continuously using AI assistance. Designers' role shifts from gatekeeper to guide. Rather than saying "no, build it this way," designers explain why—extracting principles, pointing to design systems, sharing research insights. They're embedded with engineering teams, reviewing early builds, providing real-time feedback, and often jumping into code themselves to polish details.
Wen spends significant time doing "last-mile implementation"—working directly with engineers in VS Code, tweaking CSS, refining interactions. This requires designers to be comfortable with code, not necessarily as programmers, but as collaborators who can make quick edits rather than writing lengthy specifications.
Strategic Vision and Direction: This is the hardest work to protect time for, and ironically, it's become more important, not less. In a world where anyone can build anything quickly, someone still needs to answer: What should we build? What direction ensures all these rapid experiments cohere into a coherent product?
But this vision has fundamentally changed shape. Instead of 10-year strategic decks with beautiful storytelling, today's vision is a 3-6 month prototype that points teams in a direction. It's functional, specific enough to guide decisions, but humble enough to change as reality emerges. The goal isn't to predict the future perfectly; it's to create alignment so teams aren't building in random directions.
Inside the AI-Native Design Studio
Working at Anthropic, Wen has a front-row seat to how design operates when AI capabilities are central to product development. Her time allocation tells the story:
Years Ago: 60-70% mocking and prototyping, 20% collaborative work with engineers, 10% coordination.
Today: 30-40% mocking and prototyping, 30-40% hands-on pairing with engineers, meaningful slice dedicated to direct implementation and shipping.
Her actual toolkit reflects this evolution. She uses Claude Chat for quick interactions, Claude Cowork for longer-running tasks, and Claude Code embedded in VS Code for direct implementation. She still uses Figma extensively, but for different purposes—not as the primary design source of truth, but as an exploration tool.
"Figma is still important for exploring multiple directions," she explains. "Right now, coding tools are super linear; you invest in one direction and iterate on it. Figma lets you throw ideas at the wall and curate them. It's also great for visual and interaction details that benefit from micro-exploration—typography, spacing, style variations."
This isn't either/or; it's sequential. Designers diverge and explore in Figma, then converge on a direction and move to code for implementation and refinement.
The AI Stack for Modern Designers
Wen's daily toolkit has become remarkably AI-integrated:
Claude Chat and Cowork: She shifted all chat use cases to Cowork specifically because most of her requests are longer-running tasks that benefit from sustained context and iterative refinement.
Claude Code in VS Code: Her primary tool for implementation work. Being able to see code while conversing with Claude beats typing back-and-forth specifications. She also experiments with remote Claude Code through mobile and Slack—@mentioning Claude to fix an icon, then picking up the PR when it's done.
Figma: Still essential for exploration and direction-setting, particularly for visual refinement and design system work.
Internal Infrastructure: At Anthropic, she spends significant time monitoring internal Slack channels, research updates, and prototype announcements. This isn't busywork; it's competitive intelligence. Spotting "illegible ideas"—prototypes and concepts that aren't fully understood yet but have energy around them—is part of her strategic role.
Three Designer Archetypes for the AI Era
When hiring, Wen looks for specific profiles, each valuable for different reasons.
Strong Generalists ("Block-Shaped"): Designers who are 80th percentile good at multiple core skills—not just surface-level generalists, but people genuinely strong in several areas. This is rare and hard to hire, but invaluable because design roles are stretching in multiple directions. If you're already strong at product thinking, visual design, and engineering, you can flex into whatever your role requires. With AI accelerating change, this adaptability is premium.
Deep Specialists (Extended T-Shape): Top 10% practitioners with exceptional depth in one area—whether that's technical design (practically a software engineer), visual refinement, interaction design, or animation. As Wen notes, "When everybody can make anything, having a deep specialist slant helps differentiate what we're building." When AI can generate baseline interfaces quickly, the human differentiator is often the deeply skilled craftsperson.
Craft New Grads: This is the archetype most companies overlook. Early-career designers who are humble, eager to learn, and unburdened by established processes. They don't carry baggage about "the right way to design"; they're willing to try new tools, methods, and approaches. They see possibility rather than constraint. As industries shift, people without deep investment in the old way often adapt faster.
Wen's advice to new designers is direct: "Build a bunch of stuff. Build actual things." Don't wait for the perfect brief or years of experience. Use the technology available, create real projects, find communities of makers, and share your work. The designers who will thrive in 2025 are those actively building and shipping, not those studying design theory.
Why Traditional Mocking Doesn't Work for AI Products
There's a unique challenge with designing around AI: non-determinism. You can't mock every possible state of an AI model's behavior. You can't fully predict how users will interact with a tool that generates variable outputs based on their unique inputs.
Traditional design process assumes deterministic interactions: if the user clicks here, this specific thing happens. With AI, the output space is vast. People find unexpected uses. Co-work evolved from unforeseen user interactions. Claude Code surprised everyone with how people adapted it.
"You need to use the actual models and observe people trying them out with their unique use cases," Wen explains. "You discover the true potential and applications as users interact with them." This forces a different design approach: get something functional in front of real users quickly, observe how they actually use it (which will surprise you), then iterate based on reality rather than hypothesis.
This is why at Anthropic, features like Co-work were labeled "research previews." Not as cop-out, but as honest signal: "This is powerful, we believe in it, but it's early and imperfect. Use it, give us feedback, and we'll iterate based on what you discover."
The New Designer-Engineer Dynamic
The relationship between designers and engineers is inverting in some ways. Designers aren't less important; they're differently important. Instead of handing off finished designs for implementation, they're embedded in the creation process.
The key is not trying to slow engineers down or be gatekeepers, but helping them move in coherent directions. This means:
Explaining Principles Over Dictating Solutions: Rather than "no, do it this way," Wen shares why she's thinking about something, pointing to research, user feedback, or design system principles. This equips engineers with decision-making frameworks they can apply when building without designer involvement.
Leveraging Design Systems: With Claude writing code, it's not always picking up design system patterns automatically. Pointing engineers to these resources—rather than rebuilding design specifications—is high-leverage.
Protecting Time for Coherence: In a world where anyone can build anything, design's job is ensuring all these rapid experiments feel like they belong to the same product. This is less about controlling every detail and more about setting constraints and direction that guide distributed decision-making.
Managing Craft and Quality at Velocity
One of the most compelling insights from Wen is how to maintain quality and trust when shipping constantly. The answer isn't perfectionism; it's speed plus honesty plus follow-through.
When you launch something early—whether it's Co-work or Claude Code—you make an implicit promise to users: we will iterate based on your feedback. You're building trust through speed (we fixed that yesterday) combined with visible responsiveness (we heard you, here's what we're doing about it).
"Building trust through speed" means Anthropic team members actively engage on social media when people report issues, then ship fixes visibly and quickly. This creates a feedback loop: early users see their input matters, they engage more, the team gets better data, iteration accelerates.
This is the opposite of the waterfall mindset where you polish everything privately before launch. It requires confidence in your team's ability to improve quickly, and honesty with users that early features are evolving. But it actually strengthens brand trust more than false perfection.
The key is never releasing something and then abandoning it. That's what erodes trust. Release early, commit to iterating, and follow through visibly.
Time Management for Design Leaders
Wen has an unconventional take on "low-leverage" work that's worth examining. The conventional wisdom says: as a leader, focus on high-leverage activities—only do things that only you can do. Delegate the rest.
But Wen has observed that the leaders she respects most actually choose to do seemingly "low-leverage" tasks themselves. A senior engineer submitting pull requests. A leader dogfooding the product meticulously, finding bugs, reporting them. Someone hand-coding a personalized thank you note rather than delegating it.
These activities are actually incredibly high-leverage because of who's doing them. They signal that the leader cares deeply, understands the product intimately, and isn't above any task. They create culture where quality and attention to detail matter. They build trust and psychological safety.
"It fosters a deep familiarity with the product and creates a culture where the team feels their leader genuinely cares," Wen explains. The manager's choice to do granular work becomes high-leverage through the message it sends and the culture it builds, even if the specific task could be delegated.
The Illegibility Framework: Spotting Tomorrow's Opportunities
One framework Wen finds invaluable is the "legibility matrix" from Evan Tana at SPC. It maps ideas and founders on a 2x2: legible/illegible on both axes.
When both the idea and the founder are legible, the idea is probably not novel—someone's already implementing it. But when the idea is illegible—on the frontier, not yet understood, or poorly articulated—that's where interesting opportunities hide.
Wen applies this internally, spending time in Slack reviewing prototypes and research projects, looking for "illegible ideas" with energy around them. She's not trying to fully understand them; she's spotting which early-stage explorations have momentum and might become important.
Co-work is a perfect example. There was an internal prototype called Claude Studio—a dense, powerful interface showing Claude's plans, context, and outputs. Wen didn't fully grasp it initially, but she saw energy around it from researchers and engineers. By staying curious about that "illegible" prototype and asking why people found it valuable, she identified patterns (showing plans, to-dos, context) that ultimately shaped Co-work's interface.
This is designers functioning like VCs internally—not trying to understand everything, but spotting which illegible ideas are worth investigating deeper and potentially shepherding toward the surface.
Why Design Managers Should Return to IC Work
Wen spent time as a manager at Figma overseeing 12-15 designers, then deliberately returned to full-time IC work at Anthropic. She argues this isn't a regression; it's essential.
"If I manage a team again, I'll do it with empathy and understanding of how the design process has changed," she explains. "It's difficult to empathize if you're not actively involved in hands-on work, testing tools, and trying new things."
She learned hard skills as an IC this past year that would have been impossible while managing—not just technical skills, but understanding how design actually works in an AI-native environment, what tools are valuable, where the friction points are.
This mirrors how engineering disciplines treat it: engineering managers often do rotation work to stay current with technology before managing full-time. Design should adopt similar practices, especially now when the discipline is shifting so rapidly. A design manager removed from day-to-day work for three years will be out of touch with how the role is evolving.
The hardest part of returning to IC work? "Doing crits and getting criticized," Wen admits. As a manager, you're removed from that vulnerability. Returning to regular critical feedback on your work requires re-acclimating to that exposure—but it's essential for growth.
The Legibility and Psychological Safety Dynamic
Wen manages with what she calls "tough parent" energy—caring deeply about people while holding high standards. This creates an environment where team members feel psychological safety (they won't be fired for disagreeing) while also knowing the bar is genuinely high.
This sounds contradictory but actually works powerfully together. When people feel safe, you can apply rigorous standards without defensiveness. Feedback isn't perceived as threatening; it's perceived as coaching from someone who believes in you.
Evidence of this in practice: her teams feel comfortable playfully roasting her. They mimic her phrases ("Okay, what are next steps?"). This teasing only happens in environments of genuine trust. She makes explicit that she has high standards while also making clear she's not going to fire people on a whim and is genuinely invested in their growth.
The framework is less about being friends and more about being genuinely, visibly human while also being serious about quality.
What Designers Actually Need to Learn Now
For mid-career and senior designers asking whether they need to learn coding, Wen's answer is nuanced: not necessarily learn to code from scratch, but absolutely develop fluency with AI coding tools.
"Any designer should know how to use the tools at hand," she advises, "as opposed to learning React from scratch." The point isn't becoming a software engineer; it's being able to work in an AI-native environment where you can collaborate with Claude Code, make direct edits, and iterate rapidly.
As models improve and abstractions layer higher, designers might not need to understand every line of code. But right now, familiarity with coding tools is essential vocabulary for the role.
For new grads, the path is clearer: build constantly. Ship projects. Find communities of makers. Don't wait for a job to start learning; learn by making things real and sharing them. The designers winning right now are those who've spent the last year actively building, not studying.
The Question of AI Design Quality
The final question haunting design: will AI become good enough at design that designers become obsolete?
Wen's assessment: AI will get better at taste and judgment, probably much better. But "someone has to decide what gets built and what matters." As she notes, many of the hardest parts of building software aren't actually building—they're deciding what should be built when teammates disagree, prioritizing competing visions, and being accountable for those decisions.
AI can weigh in on aesthetic questions. It might eventually make good taste calls. But the meta-decision of what the product should be—that still requires human judgment and accountability. Just as engineers remain accountable for code quality even though AI writes most of the code, someone needs to be accountable for product decisions.
This doesn't mean designers are safe by doing the same work. It means the valuable work is shifting toward judgment, vision-setting, and accountability rather than pixel perfection and detailed specifications.
Hiring for the Frontier
Wen's team at Anthropic is actively hiring designers across three archetypes: strong generalists, deep specialists, and craft new grads. If this describes you and you're excited about frontier work, she's looking for:
- Genuine enthusiasm for AI technology and how it's reshaping products
- Track record of building things (portfolio matters less than evidence of shipped work)
- Resilience and willingness to adapt when things change rapidly
- Either breadth across multiple skills or exceptional depth in one area
- Comfort with ambiguity and iteration
The hiring bar isn't about prior experience at prestigious companies or a perfect portfolio. It's about demonstrating you can navigate rapid change, learn new tools, and ship work that matters.
The Design Stack in 2025
For designers adapting to this new reality right now, Wen's stack offers a practical roadmap:
For Exploration: Figma remains valuable for diverging on multiple directions, exploring micro-variations, and visual refinement.
For Implementation: Claude Code (or similar AI assistance) embedded in your IDE for direct code collaboration and refinement.
For Long-form Work: Cowork or similar tools for sustained, context-heavy projects that benefit from extended conversation.
For Strategic Thinking: Time in internal communication channels, prototype reviews, and spotting illegible ideas that might become important.
For Execution: Willingness to write code yourself, even if not at a software engineer level, to refine and polish details.
This isn't a either/or stack; it's a sequential workflow where different tools serve different purposes in the design process.
The Broader Industry Shift
While Wen's observations come from working at the frontier at Anthropic, she notes this shift is rippling across the industry. Her talk about design process becoming obsolete resonated broadly—though it also generated backlash from designers who've built careers around traditional methodology.
"There is still a piece of the industry not quite there yet," she acknowledges. Some teams absolutely still benefit from deep discovery and traditional process. But she sees emerging evidence across companies at various maturity levels that something is shifting—people are using Claude Code and similar tools to prototype faster, PMs are spinning up prototypes themselves, iteration velocity is increasing.
The question isn't whether these changes will happen; it's whether your team will adapt deliberately or fight change until forced to.
Designing With Non-Deterministic AI
Because AI models behave non-deterministically, traditional design approaches fail. You can't mock every possible state. You can't prototype every interaction. You have to ship something real and let users show you how they'll actually use it.
This requires a different relationship with "quality." Shipping a research preview that's 80% polished but reveals genuine value is better than shipping nothing in pursuit of 100% perfection. The promise you make is that you'll iterate based on feedback and actually follow through.
At Anthropic, this means labeling features like Co-work as "research previews," being transparent about rough edges, but confidently shipping because the core value is real. Then the team stays engaged, responds to feedback publicly, and ships improvements visibly and quickly.
This approach maintains trust better than waterfall perfectionism because users see their feedback matters and the product actually improves based on their input.
The Future of Design: Neither Dead Nor Unchanged
The traditional design process is dead in the sense that elaborate mocking and multi-month refinement before shipping no longer matches how products actually get built. But design itself is far from dead—it's being redefined.
The designers thriving right now are those who've embraced the shift: getting comfortable with implementation, moving from specification to collaboration, learning to work in AI-native environments, and protecting time for strategic direction even as execution velocity accelerates.
The designers struggling are those waiting for the industry to come back to them, hoping that "good process" will return as a valued activity. It won't. But directing the flood of rapid development, maintaining craft through velocity, and deciding what actually matters to build—that work is more valuable than ever.
Design isn't dead. The old design process is. And that might be the best thing that could happen to the profession.
결론
The design profession stands at an inflection point. The tools, timelines, and structures that defined design for decades are being replaced by something faster, more collaborative, and less about perfect specification and more about strategic direction.
For designers starting careers, the path is clear: build constantly, learn AI tools, embrace ambiguity, and move quickly. For mid-career designers, this might require learning new tools and rethinking your relationship to code. For design leaders, it means protecting time for vision while staying hands-on with execution and embedded with your teams.
The designers who will thrive in 2025 and beyond are those who see this moment not as threat, but as liberation. You're no longer gatekeeping perfect designs; you're directing the flow of rapid creation, maintaining coherence across distributed teams, and deciding what actually matters to build. That's harder work than elaborate mocking, and ultimately more valuable.
The future of design isn't AI replacing designers. It's designers who embrace AI, speed, and a redefined role becoming more essential than ever to building products that actually work at the frontier.
Original source: The design process is dead. Here’s what’s replacing it. | Jenny Wen (head of design at Claude)
powered by osmu.app