Design Process Is Dead: Anthropic's Design Lead Jenny Wen on the AI-Era Design Shift设计流程已死：Anthropic 设计负责人 Jenny Wen 谈 AI 时代的设计变革

Jenny Wen is Anthropic's Head of Design for Claude, currently leading design for Claude Co-work. Before joining Anthropic, she was a Design Director at Figma, leading FigJam and Slides from concept to launch. Earlier still, she worked in design at Dropbox, Square, and Shopify.

This episode of Lenny's Podcast covers five big topics: why the traditional design process has reached its end, what it feels like to do design at an AI lab, whether AI will replace human taste and judgment, how Co-work actually got built, and what kind of designers are being hired now.

Source: Lenny's Podcast, March 1, 2026. Original video: https://www.youtube.com/watch?v=eh8bcBIAAFo

Key takeaways

The traditional design process is "dead": the classic diverge-converge-diverge-converge process has been forced out by engineering speed — the time designers spend making mockups has shrunk from 60-70% to 30-40%, with the freed-up time spent pairing with engineers or even writing code themselves

Vision planning has shrunk dramatically: the time horizon has gone from 2-5 years down to 3-6 months, and the format has shifted from polished slide decks to prototypes that simply point a direction

Human value lies in decisions and accountability: AI will keep getting better at taste and judgment, but the hardest part of building software isn't the building itself

The real story behind Co-work: the "10 days" of development was actually the final sprint after a long period of exploration, and brand trust doesn't come from a perfect launch but from responding quickly to feedback

Three types of designers most worth hiring: strong generalist "squares," deep T-shaped specialists, and craft-minded new grads — the third type is the most overlooked but the most valuable during a period of transformation

The design process didn't die on its own — it was killed off by engineering speed

Lenny's opening question: how has the design process changed in the AI era?

Jenny's answer was direct:

The design process designers were taught — we used to treat it almost like gospel — is basically dead now. ("This design process that designers have been taught—we sort of treat it as gospel—that's basically dead.")

She's referring to the classic double-diamond model: research and diverge, then converge, then diverge again, then converge again.

This methodology was already struggling to hold up even before AI, but once engineers could run seven Claude instances at once to build features, designers simply couldn't keep working with the old process at all.

Jenny gave a talk called "Don't Trust the Design Process" at the Hatch Conference in Berlin in September 2025, and it got a huge response. But barely three or four months later, she already feels the content is out of date. Especially after the release of Opus 4.6 and the wave of people discovering Claude Code over the holidays, the design process has changed even faster than she expected.

She now splits design work into two categories. The first is supporting execution: engineers are shipping at high speed, and anyone can pitch an idea and have an engineer (or an AI) knock out a rough version to try, with the designer acting more as an advisor than someone who draws mockups first and hands them off. The second is setting vision and direction, but its form has changed too: in the past you could produce a 2-year or even 10-year design vision as a beautifully told slide deck; now the vision usually only extends 3-6 months out, and sometimes takes the form of just a prototype that points a direction.

You're better off not blocking that, letting them cook. ("You're better off not blocking that, letting them cook.")

Lenny followed up: is this shift happening across all companies, or only at AI labs?

Jenny said the response to her Berlin talk was stronger than she expected. Product managers are prototyping with Claude Code, and designers are developing with v0. But there's also plenty of pushback: some designers have built their entire careers around this process, and they're not willing to accept the idea that "we can skip discovery research."

On "ship fast versus polish carefully," Jenny thinks it depends on the situation. But there's a fundamental reason why rapid iteration matters especially for AI products: AI models are non-deterministic — you can't simulate every state in a mockup, and you can't even build a meaningfully clickable prototype. You have to use the real model and watch real users to discover the actual use cases.

[Note: Non-deterministic means the same input can produce different outputs. The traditional approach of "mocking up every screen state" breaks down for AI products, because the AI's responses themselves are unpredictable.]

A day in the life of a designer at Anthropic

Lenny asked a very practical question: what does a designer at an AI lab actually do day to day?

Jenny said she spends a fair amount of time just "keeping up." At Anthropic, at any given moment, many teams are prototyping, testing new ideas, and pushing forward various codenamed projects.

Our Slack is a gold mine. ("Our Slack is a gold mine.")

From progress on model capabilities to internal debates about where the industry is headed, she tries to keep up with all of it. This information is directly useful to her work: she needs to anticipate what might come next so she can prepare for it design-wise ahead of time.

Beyond staying informed, Jenny's day roughly breaks down like this: some time reserved for traditional design thinking; a large amount of time spent bouncing ideas off engineers — talking, whiteboarding, looking at what they've built, giving feedback; and some time spent directly polishing things in code.

On how her time allocation has shifted, she gave clear numbers:

A few years ago: 60-70% on mockups and prototypes, 20% collaborating with engineers, 10% in coordination meetings

Now: 30-40% on mockups and prototypes, 30-40% working directly with engineers, plus a new chunk — writing implementation code herself

When working with engineers, her focus is on explaining "why." Instead of saying "the button shouldn't go here," she'll say "I think there should be a button here, because user research shows not everyone knows this feature can be triggered with a prompt." She also tries to steer engineers toward using existing components from the design system, since Claude doesn't always automatically use the design system when writing code.

Her AI toolkit: Claude Chat has basically been replaced by Claude Co-work, since most of her use cases are long-running tasks. Claude Code is mainly used inside VS Code, since polishing the front end requires looking at code and talking to Claude at the same time. One workflow she finds especially fun: someone in Slack says "this icon looks off," tags Claude, Claude automatically fixes the code and commits it, and she just merges it — done.

[Note: Claude Co-work is a desktop AI agent product Anthropic launched in January 2026 that can operate on files on a user's computer, handling non-coding knowledge work like document generation and data organization.]

Is Figma still useful?

Given Jenny's Figma background, Lenny asked directly the question a lot of people care about.

Jenny said she still uses it, and believes Figma remains important — but for different reasons than before.

The problem with code-based tools is that they're too linear. If you use Claude Code to pursue one direction, you end up iterating deeper and deeper along that same direction. But good design requires first coming up with 8-10 different approaches, throwing a bunch of ideas at the wall, then filtering and pushing yourself to explore further possibilities. For this kind of divergent exploration, Figma's canvas is still the best tool available.

Another value is fine-grained visual tuning. Comparing different layouts, fonts, and style directions side by side on a canvas is far more efficient than repeatedly switching back and forth in code.

Lenny pointed out an interesting phenomenon: in engineering, the IDE is being displaced by the command line and agents, and engineers no longer think IDEs are "cool." But for designers, the IDE has actually become a useful tool, because sometimes directly tweaking a CSS style is faster than describing it to Claude. Perhaps the IDE is becoming a tool for designers and product managers, while engineers have already moved on.

The hardest part of building software isn't building it

Citing something Lex Fridman said, Lenny asked Jenny: as AI gets smarter, where does the human brain still add value?

He brought up something Claude Code lead Boris Cherny recently said on the show: Claude Code is no longer just writing code — it's starting to help him come up with ideas and decide what to work on. That made Lenny reconsider the assumption that "AI will never make judgment calls the way a good product manager or designer does."

[Note: Boris Cherny is the creator and lead of Claude Code. On Lenny's Podcast in February 2026, he said "coding is basically already solved," and that Claude Code is now starting to scan feedback, bug reports, and telemetry data to proactively suggest improvements.]

Jenny believes AI will keep getting better at taste and judgment — "we may be too fixated on that point." But she pointed to a more fundamental issue:

A lot of the hard parts of building software are actually, like, not building it. ("A lot of the hard parts of building software are actually, like, not building it.")

Think back to the hardest moments in your work — they're usually not about technical implementation, but about arguing with another person over "should this feature even exist" or "what should it actually look like." AI can offer input on these interpersonal decision disputes, but it can't resolve them for you.

Just as Claude can now help engineers write code, engineers are still accountable for whether that code is correct and whether it belongs in the product. The same goes for design and product decisions — the decisions and the accountability still rest with people.

Lenny added an analogy from radiology: AI might be better at diagnosis than a radiologist, but you still need a human to sign off, because someone has to be accountable if something goes wrong.

Jenny also admitted that we may be underestimating how fast AI is improving at these things.

Chat or graphical interface

Lenny noted that nobody expected chatbots and terminals to become AI's lasting interface, yet not only have they not disappeared, they've become even more dominant.

Jenny thinks the future will combine both: clickable graphical interfaces plus conversation. Claude recently released a series of widgets (weather, stocks, multiple-choice questions, etc.) that have gotten a great response, because people still like seeing a UI, clicking on it, and interacting with it — that's far more efficient than typing.

But chat as a paradigm opened a huge door: it gives you countless ways to communicate with a computer. So chat won't disappear, but for specific tasks, a UI is still more direct. The likely trend going forward is that more and more UI will be dynamically generated by the model itself, rather than hand-built one piece at a time by engineers.

Lenny brought up a point from Kevin Weil: language is an interface that works across every level of intelligence — you can chat with someone with an IQ of 200, and you can chat with someone less sharp, and language works for both. So as models get smarter, conversation will keep working.

[Note: Kevin Weil is OpenAI's Chief Product Officer, and previously held executive roles at Instagram and Twitter.]

From director back to IC: what this year taught me

At Figma, Jenny managed a design team of 12-15 people plus a few design managers — a proper design director role. But when she joined Anthropic, she chose to become an IC (Individual Contributor).

On one hand, she wanted to personally feel the changes in tools and process firsthand during the AI era. On the other, she has real anxiety about the future of middle management — as AI reshapes how work gets done, will management roles even continue to exist?

Over this year at Anthropic (starting as an IC, briefly managing a team for a few months, then returning to being an IC), she feels she's gained enormously. The design process has changed so fast over the past year that if she'd stayed purely in management, she simply wouldn't have had time to pick up these new hard skills. If she manages a team again in the future, this experience will let her genuinely understand the challenges her team faces, rather than just guessing from the outside.

She recommends that design managers should also do something like the "hands-on rotation" that engineering managers do — spend a few months as an IC to understand the technical shifts firsthand, then go back to managing.

Lenny asked what she found hardest to readjust to after going back to being an IC. Jenny laughed and said: taking criticism. As a designer, presenting your work to the team and receiving critical feedback is a fairly vulnerable process, and you get rusty at it after spending a long time in management.

On the future of management, Jenny believes that as long as there are teams, there will need to be managers. But future managers will need to both set direction for the team and do a share of IC work themselves — pure "people management" as a standalone role may no longer be enough.

The real story behind Co-work

Boris Cherny said on Lenny's show that Co-work was built in 10 days, and that figure spread widely online. Lenny asked Jenny what actually happened.

Jenny corrected the impression: the 10 days was the sprint from internal version to external launch. Before that, the team had done extensive prototyping and exploration across different agent frameworks — testing many different approaches for how to display the to-do list, what format to use for multiple-choice questions, and how to teach users to understand the use cases.

This idea kept resurfacing, and then suddenly, the timing was right, and it felt like it had always been obvious. But the journey to get there was long, very long.

("The idea kept coming back, and then all of a sudden, it's the right moment, and it feels like it was so obvious all along. But there was a long, long journey to get there.")

On launch strategy, Jenny said Co-work wasn't perfect when it launched, but the team had used it extensively internally and was confident it had real value worth letting external users experience. The key is: after launch, you have to follow through on your promises.

What really damages a brand is releasing an early version and then doing nothing. ("The way that you really lose trust around quality... is if you release it early and then nothing ever happens.")

Lenny summarized this philosophy as "building trust through speed." Jenny added that it's not just about speed, but also making users feel that "my feedback was heard and acted on." After every new release, Anthropic team members reply to user feedback on Twitter, fix issues quickly, and publicly show progress.

Lenny asked what she's proudest of about Co-work. Jenny said she's proudest that they shipped it. Because when you're the designer, you only ever see the flaws in your own work.

Lenny asked how Co-work should be described in one sentence. His own phrase was "Claude with hands." Jenny said she likes that, but her own description is more down-to-earth: what Co-work is good at is you throwing it a pile of messy stuff, and it turns that into something neat and useful.

Her current directions for iteration:

Making Co-work's homepage feel more like a shared task list between you and Claude

Thinking about whether Co-work will always live only on a screen, or whether it could extend to other work surfaces

The three types of designers she most wants to hire

Lenny asked what she looks for when hiring designers in an era when everything is changing.

Jenny said the first thing is resilience and adaptability — being willing to try new methods, learn new tools, and not cling to old processes.

More specifically, she's currently most interested in three types of people:

The first: the "square"-shaped strong generalist. Not someone who dabbles in everything without going deep, but someone who's at the 80th percentile across multiple dimensions. A traditional T-shaped person is deep in one area and shallow elsewhere; a square-shaped person is deep in several directions at once. This kind of person is especially valuable in an era when role boundaries are blurring, as design work is increasingly extending toward product management and engineering. Jenny also admits this type of person is rare.

The second: the deep T-shaped expert. The vertical stroke of the T is much longer than most people's, ranking in the top 10% of the industry in some domain. This could be a designer who's extremely technical, essentially half an engineer, or it could be someone who's a top expert in visual design or icon design. When everyone can use AI to produce something "okay," it's deep expertise that creates differentiation.

The third: a craftsmanship-minded new graduate. Early in their career, but with maturity beyond their age, quick to learn, without fixed process-oriented thinking. Most companies are competing for senior talent, but precisely because the rules are changing, a blank-slate fast learner may have an advantage over a senior person whose head is full of old processes.

Advice for young designers: make lots of things, and don't let "lack of experience" hold you back. Jenny mentioned Socratica, a community at her alma mater, the University of Waterloo — a student maker community that meets weekly to work together in person, build projects, and show them off. Someone built a Claude-powered robot; someone stuck cartoon eyes onto buses in Boston. That kind of "I'm just going to go make something" drive is what makes people stand out.

[Note: Socratica is a student community founded at the University of Waterloo in 2022, now expanded to more than 30 cities worldwide.]

On the question of "should designers learn to code," Jenny's advice is practical: you don't need to learn React from scratch, but you should bring AI coding tools into your toolkit. As models and products get better, the layer of abstraction will keep rising, and designers won't need to understand how every line of code runs.

Lenny asked a pointed question: how good is Claude as a designer? Would you hire it?

Jenny was direct: not good enough yet. Claude doesn't fit any of the three archetypes she mentioned — it's fine at drafting and showing different options, but there's nothing that makes you feel "this is special, worth hiring for." That said, she also noted that Claude has improved a lot on this front over the past year.

The counterintuitive wisdom of management

The second half of the interview turned to team management. Jenny shared several interesting perspectives.

Low-leverage time

Management training teaches you to sort work using a 2x2 matrix — "only I can do this" versus "others can do this too" — and then cut out all the "low-leverage" tasks. But Jenny has observed that the leaders she respects most often deliberately choose to do some "low-leverage" things — and precisely because it's them doing it, those things become high-leverage.

For example, executives who personally spend a lot of time testing the product, reproducing bugs, and digging into logs alongside engineers. A leader doing this personally builds deep familiarity with the product, and it also sends the team a signal that no task is beneath anyone. Mike Krieger personally submitting code is one example. Another example: a leader personally making a carefully designed keepsake card for an employee — admin staff could do this task, but a leader doing it themselves sends a completely different message.

[Note: Mike Krieger is a co-founder of Instagram. He joined Anthropic as Chief Product Officer in 2024, and moved to the Anthropic Labs team in early 2026.]

A culture of ribbing each other

When team members are willing to joke with each other — even daring to joke about their manager — it shows they aren't afraid of you and trust you. People on Jenny's former team used to imitate her catchphrase from design reviews, "OK, what's next?" — which showed they knew her and weren't afraid of her.

But this has to coexist with high standards. She uses the metaphor of "strict parents": the team knows you won't fire them on a whim, but also knows you expect the best work. With psychological safety as the foundation, it actually becomes easier to hold high standards. Lenny summarized this as the classic formula of Radical Candor: deep care combined with direct challenge.

The legibility matrix

The third topic came from Evan Tana's "Legibility Framework." The matrix has two axes: whether the founder is "legible" (easy for others to understand at a glance), and whether the idea is legible. If both the founder and the idea are highly legible, that opportunity is probably already being pursued by someone else. The most valuable quadrant is often the one where the idea is illegible — something others can't quite understand, but where energy is gathering.

[Note: Evan Tana is a partner at SPC (South Park Commons, a Silicon Valley startup community and fund).]

Jenny applies this framework in her daily work: when she browses various internal prototypes on Anthropic's Slack, she's looking for things that are "illegible" but have energy behind them.

One concrete example. Last year, someone inside Anthropic built a prototype called "Claude Studio," with an interface that was extremely dense and complex, built on some kind of agent framework. Jenny's first reaction was "I don't know what this is." But she noticed the research team and internal users were very excited about it. Instead of ignoring that signal, she chose to dig deeper. Eventually, the core concepts from that prototype — such as the Skills framework (using Markdown files to guide Claude on how to complete specific tasks), and the UI showing Claude's plans and to-do items — were extracted and incorporated into Co-work's design.

Lenny added a related finding: his research with venture capitalist Terrence Rohan shows that people who joined companies early that later became huge successes (like Palantir, Stripe, Linear, OpenAI) saw three signals: the idea sounded crazy, some people were extremely excited about it, and the founder was top-1%-caliber talent.

Jenny said this matches her own experience: when you see something you don't understand but that people are excitedly throwing themselves into, it's worth digging deeper. Early-stage creators often can't articulate why they're excited, and someone needs to help translate that vague energy into a clear product.

Lightning round

Recommended book: The Power Broker (by Robert Caro, about the life of Robert Moses), 1,100 pages. Jenny said that in an age of scarce attention, reading a biography spanning decades is especially valuable. The other is Insomniac City (by Bill Hayes), a memoir about the final period of scientist Oliver Sacks's life.

Recently enjoyed film: A Sentimental Value, a new film by Norwegian director Joachim Trier (who also directed The Worst Person in the World), about a family's relationship with the house they've lived in their whole lives. Also, The Pit season two — watching extremely capable people do what they're good at is just enjoyable to watch.

Favorite product: Retro, a small-circle photo-sharing app where you can only share photos from the current week, with none of the counting and advertising typical of social media. After using it for two years, you can look back at "what I was doing this week two years ago," which has become a way of recording life.

Life motto: "It is what it is." It sounds like resignation, but Jenny says that in a world where everything is changing, this phrase gives you the ease you need to keep moving forward.

Coolest use of Co-work: Jenny fed years of her own notes — one-on-one notes, random thoughts, small memos, interview notes — all into Co-work and had it analyze what she values when assessing design craft. The output was a set of evaluation criteria that she herself hadn't consciously realized. When AI can help you discover your own implicit patterns of thinking, that in itself is valuable.

The whole podcast episode has just one central thread: change is not being initiated from within the design world itself, but rather the explosive growth in engineering efficiency is pushing designers into a position where they must change. Designers need to shift from being gatekeepers of process to being guides, from people who draw mockups to people who can do fine-tuning directly in code.

One signal worth watching is Jenny's mention of Co-work's next step: "whether it will always live only on a screen." This hints that Anthropic may be exploring ways for AI agents to reach more work surfaces, rather than cramming all interaction into a single chat window.

Another open question is how fast AI's taste and judgment will evolve. Jenny admits Claude currently isn't good enough to be hired as a designer, but she also says it has "improved a lot over the past year." That gap is narrowing, and no one knows how much it needs to narrow before it triggers another shift in the industry.

Anthropic's design team is hiring. If the idea that "the design process is dead" excites you rather than scares you, Jenny says: welcome.

Full interview video: https://www.youtube.com/watch?v=eh8bcBIAAFo