Jenny Wen: The Design Process Is Dead, and Designers Are Holding On Too Tight

Jenny Wen left a Director of Design role at Figma to become an IC at Anthropic. Not because she was done with management, but because she judged this technological moment significant enough to warrant getting her hands dirty again. A year in, the design process she was taught and once defended is, in her words, “basically dead.” What replaces it is messier, faster, and more honest about where human value actually lives.

The Episode

Jenny Wen joins Lenny Rachitsky on Lenny’s Podcast. She leads design for Claude Co-work at Anthropic, having previously led the teams behind FigJam and Slides at Figma. The conversation spans 77 minutes and covers why the traditional design process collapsed, what a designer’s day actually looks like inside an AI lab, whether AI will develop taste, why chatbot interfaces have more staying power than expected, the decision to step down from director to IC, three hiring archetypes for the new era, and a framework for spotting illegible ideas before they become obvious.

The Process That Died

Jenny gave a talk in Berlin last fall called “Don’t Trust the Design Process.” The core claim: the discover-diverge-converge-iterate cycle that designers treated as gospel was already dying before AI, and AI killed what was left.

The numbers tell the story. A few years ago, 60-70% of her time went to mocking and prototyping. Now it’s 30-40%. The freed-up time didn’t disappear into meetings. It went to jamming directly with engineers, consulting on in-progress work, and writing actual code.

“This design process that designers have been taught, we sort of treat it as gospel. That’s basically dead.”

The cause isn’t that design tools changed. It’s that engineering speed exploded. When engineers can spin up seven Claude Code agents and ship features continuously, a designer who insists on producing polished mocks before anything gets built becomes a bottleneck. Jenny’s advice is blunt: “Let them cook.”

“You’re better off not blocking that, letting them cook.”

And the pressure isn’t one-directional. It’s not just designers struggling to keep up with engineers. Engineers are struggling to keep up with themselves. “How do we keep up with all our agents?” she asks. There are seven agents constantly running.

Two Kinds of Design Work Now

Jenny sees the profession splitting into two modes.

Execution support. Engineers generate a first pass with AI. Designers step in to refine, consult, give feedback at the code level, polish the last mile. This is where most of the day goes. It’s less glamorous than producing hero mocks, but it’s where quality actually happens now.

Vision and direction. Someone still needs to point the team toward something coherent. But the shape of this work has compressed dramatically. Design visions used to span 2-5 years and live in beautifully story-told decks. Now it’s a 3-6 month prototype that points people in the right direction.

“Now it becomes a vision that’s 3 to 6 months out and isn’t necessarily creating this beautiful deck. Sometimes just creating a prototype that points people in the right direction.”

The second kind of work matters more than ever precisely because the first kind is so fast. When anyone can ship anything in any direction, someone has to ensure it all makes sense together.

Inside Anthropic: What the Day Actually Looks Like

A surprising chunk of Jenny’s time goes to just keeping up with what’s happening inside the company. Anthropic’s internal Slack is, in her words, “a gold mine.” Research developments, prototype codenames, philosophical debates about product direction. She described it as the best AI news feed in the world, except it’s internal.

Her tool stack is fully Claude: Claude chat, Claude Co-work (she’s shifted all her chat use cases to it), and Claude Code through VS Code for frontend polish. She still uses Figma, but specifically for exploratory work where you need to throw eight different directions at the wall and compare them side by side. AI coding tools are too linear for that kind of divergent exploration.

An interesting detail: she’s started using Claude Code remotely through Slack. Someone mentions a misaligned icon, she @mentions Claude, Claude fixes it, she picks up the PR. Done.

Co-work: Not 10 Days

The viral narrative around Claude Co-work was “built in 10 days.” Jenny corrects this. The 10 days were from internal prototype to shippable product. Behind that were months of exploration, multiple prototypes on different agent harnesses, and dozens of people experimenting with different interaction patterns.

One internal prototype called “Claude Studio” was dense, powerful, and illegible to most people. Jenny looked at it and didn’t get it. But she noticed the energy around it from researchers and engineers. That energy was the signal. The skills framework (markdown files that instruct Claude on specific tasks) and the todo-list-like interface in Co-work both trace their lineage to that prototype.

Her proudest moment with Co-work isn’t a specific feature. It’s that they shipped it. The promise of “research preview” is: we’ll put it out early and keep iterating based on your feedback. Trust comes from speed, not from perfection at launch.

“The way that you really lose trust is if you release something early and then nothing ever happens.”

Will AI Develop Taste?

Lenny brought up Boris Cherny’s claim that Claude Code is already helping him come up with ideas. Jenny’s response was more candid than most designers would be comfortable with.

“I think we might be holding on to that a little bit too much.”

She thinks AI’s sense of taste will improve, and designers may be overvaluing taste-as-moat. But she draws a line: even as AI gets better at generating options, someone still has to decide what gets built. And the hardest parts of building software were never the building. They’re the disagreements. Two people arguing about what should go into a feature. AI can weigh in, but it can’t resolve the dispute.

“At the end of the day, someone has to decide what is actually going to get built and what actually matters. Someone still needs to be accountable for the decision.”

The radiology analogy came up: AI might diagnose better than humans, but someone still needs to sign off. That’s not the most exciting job description, but accountability can’t be automated.

Why Chatbots Won’t Die

Against the popular narrative that chat interfaces are a temporary stop, Jenny argues they have real staying power.

The reasoning is anthropological. Talking is how humans have communicated for millennia. It scales across every level of intelligence. Kevin Weil made this point on the podcast: you can talk to someone at 200 IQ or 300 IQ and talking still works. It’s a universal interface.

Chat will coexist with richer UIs. Anthropic is already shipping interactive widgets within the chat (weather, stocks, structured questions). But the chat layer isn’t going away. Co-work, OpenClaw, and Claude’s various integrations through WhatsApp, Telegram, Slack are all evidence: more surfaces for chat, not fewer.

Director to IC: A Calculated Retreat

Jenny managed 12-15 designers and several managers at Figma. She left that to become an individual contributor at Anthropic. The reasoning was practical, not romantic.

She was having doubts about whether middle management would persist. The skills she’d need in the next era were changing faster than she could learn them from a management seat. An IC rotation would give her direct experience with how AI changes design work, empathy for what her future teams would face, and hard skills she couldn’t pick up through delegation.

A year later, she says the IC stint was the best investment she made. The design process changed so much in 12 months that she wouldn’t have understood it from a management perch.

She’ll probably go back to management. But with updated intuitions. She compared it to how engineering orgs make new managers take a rotation writing code before managing engineers. Design needs the same practice.

The most surprising thing about going back to IC? Getting used to crits again. Presenting your work and hearing critical feedback regularly is a vulnerable exercise, and that muscle atrophies in management.

Three Hiring Archetypes

Jenny is hiring for Anthropic’s design team and defined three profiles she’s excited about.

The strong generalist. Not “kind of good at a lot of things” but 80th-percentile good across multiple disciplines. Block-shaped, not T-shaped. These people can flex as the role stretches into PM territory, engineering territory, research territory. Extremely rare and hard to hire.

The deep specialist. T-shaped but the stem goes deeper than most. Top 10% in the industry at something specific, like visual design, interaction design, or technical design that’s essentially half software engineering. When everyone can make anything, deep specialization is what creates differentiation.

The craft new grad. Early career, wise beyond their years, humble, eager learners with no baked-in processes to unlearn. Jenny thinks most companies are overlooking this archetype. In a world where the rules are being rewritten, a blank slate and a learning speed advantage might matter more than ten years of experience with a dying process.

Her advice to new designers: just build stuff. Don’t wait for permission or education. The best candidates she’s seen are people who use the technology, build actual things, and share them with a community.

Low-Leverage Leadership

Jenny has a contrarian take on management: the tasks everyone tells you are “low leverage” are often the highest leverage.

The canonical example: a senior leader who personally tests the product, repros bugs, shares logs with engineers, nitpicks. This is supposed to be below a director’s pay grade. But Jenny argues it creates product familiarity that makes every other decision better, and it signals to the team that nothing is beneath anyone.

She cited seeing Mike Krieger (Anthropic’s CPO) submit PRs himself. It’s powerful not because the code matters, but because it destroys the hierarchy of “leaders don’t do that.”

The Legibility Framework

Jenny shared a framework from Evan Tana (partner at SPC) that she can’t stop thinking about. Ideas and founders can each be legible or illegible. The most interesting opportunities sit where the idea is illegible: there’s energy around it, people are excited, but nobody can quite articulate why.

She connects this to her role at Anthropic. Browsing internal Slack prototypes, she’s looking for illegible ideas with energy. Claude Studio was one of these: dense, confusing to look at, but researchers couldn’t stop using it. The elements that made it compelling eventually became Co-work’s core features.

Her framing: designers at frontier labs should think like VCs. Spot the illegible ideas with energy, understand what’s really driving that energy, and figure out how to make it legible to the world.

A Few Observations

• The most important sentence Jenny said was the quietest one. “We might be holding on to that a little bit too much,” about designers clinging to taste as their moat. From someone inside Anthropic seeing model progress daily, this is less an opinion and more a field report. Designers who treat taste as an unchallengeable fortress should pay attention.
• The Director-to-IC move is a leading indicator. When someone who has already proven they can manage chooses to give up hierarchy for frontline work, it signals their private assessment of the situation. Jenny’s bet is that the skills acquired this year as an IC will be worth more than another year climbing the ladder.
• “Let them cook” is not just design advice. It’s an organizational philosophy for the AI era. The traditional handoff model (design specs → engineering builds) assumed building was the bottleneck. Now iteration is so fast that the bottleneck is direction and judgment. The role that matters most is the one that points all the autonomous agents toward something coherent.
• The 10-day Co-work story is a lesson in narrative compression. Everyone heard “10 days” and assumed it was a sprint from zero. The real story was months of illegible exploration, dead-end prototypes, and accumulated insights that suddenly crystallized when the moment was right. This pattern recurs in every “overnight success.”
• The three hiring archetypes map to three different bets about the future. Strong generalists bet that roles will keep blurring. Deep specialists bet that differentiation will matter more as AI commoditizes the baseline. Craft new grads bet that unlearning matters more than experience. Jenny wants all three, which is itself a hedge against an unpredictable future.