The design process is dead. Here’s what’s replacing it. | Jenny Wen (head of design at Claude)
These notes were generated by AI and may contain inaccuracies.
This design process that designers have been taught, treated as gospel, is basically dead. Designers no longer have time to make beautiful mocks. A big part of the design role now is helping engineers and teams execute, not just handing over designs. Few years ago, 60-70% of design work was mocking and prototyping; now it's 30-40%. Better off not blocking engineers, letting them cook. Engineers are also struggling to keep up with themselves and their agents—there are seven agents constantly running. Engineering changes force design to change. Visions used to be 2-5-10 years out with beautiful decks; now 3-6 months out, often just prototypes pointing in the right direction. Boris on the podcast said Claude Code helps him come up with ideas. Designers might be holding onto taste, judgment, and design too much. Human brains remain valuable for deciding what gets built and what matters—someone needs to be accountable.
Jenny Wen was head of design for Claude, now leading design for Claude Co-work. Previously director of design at Figma (led FigJam and Slides), designer at Dropbox, Square, Shopify. Conversation gives glimpse into future of design profession.
Engineering has changed more than design so far, forcing design to adapt. Talk in Berlin September called "Don't Trust the Design Process": traditional process (research, discovery, diverge, converge) treated as gospel is dead, dying even before AI, accelerated by engineers spinning up seven clones. Now two types of design work:
- Supporting implementation/execution: engineers use seven clones to create features; anyone can propose ideas; engineers make scrappy versions; designers lack time for beautiful mocks.
- Creating vision/direction: hardest to make time for; now 3-6 months out, not beautiful decks but prototypes; essential to point teams amid rapid feature creation for efficiency and cohesion.
Engineering speed forces designers to let engineers ship without blocking, help along, ensure cohesion. Not one unifying voice demanding change, but follow-on effects of engineering tools. Design tooling will change next year. Designers empowered by coding tools for last-mile polish, prototyping in code.
Talk resonated across industry; people feel old process unfeasible with tools like Claude Code, v0 for prototypes; PMs also prototyping. Backlash from those invested in traditional process (discovery, research). Question: discovery/mocks/iterating vs. ship good-enough, learn/iterate—execution with real data/users leads to better products, especially with non-deterministic AI models. Can't mock all states or theorize; must use actual models, discover use cases from real usage. Don't know how people will use AI or its capabilities.
Much time catching up on company happenings: model developments, teams prototyping (code names), labs closer to research. Spotting ahead, trying internal prototypes/products. Reading philosophical debates/directions. Compelled to keep up due to volume and interest. Connects to helping engineers execute, connect ideas, cohesive experience.
Slack is gold mine for internal AI news. Still traditional future-thinking: allotted time this week for next 3 months vision for Co-work given market/models. Jamming with engineers: conversations, whiteboarding, feedback, consulting. Time in code polishing/implementing with engineers—fun, new a few months ago.
Still do user research (researcher does studies/surveys, team reads), prototyping, mocking—but wider tools, proportions changed. Few years ago: 60-70% mocking/prototyping, 20% jamming/consulting engineers, 10% coordination. Now: 30-40% mocking, 30-40% jamming/pairing engineers, slice for implementation/building/shipping.
Deep on Claude stack: Claude Chat, increasingly Claude Co-work (shifted chat use cases for longer tasks), Claude Code (mostly VS Code IDE for frontend tweaks; trying remote via mobile/Slack, e.g., @mention Claude to fix icon, pick up PR). Fully Claude house. Still uses Figma.
Figma fills gap: exploring 8-10 options, throwing ideas at wall, curating—coding tools linear, invest in one direction. Fine visual/interaction details (typography, styles) in canvas. IDE useful for quick edits (CSS, hex values, classes) vs. describing to agent.
Designers struggle to keep up with engineers shipping constantly. To help engineers improve at design, explain reasoning and principles behind decisions rather than just saying no. For example, "We should have a button here because not everybody realizes you can prompt this," backed by research examples. Point engineers to the design system and code. Currently, Claude writes a lot of code but doesn't always follow the design system.
Equip engineers with tools they can use independently. Both engineers and designers feel overwhelmed by their own capabilities and desire to do more. Engineers also struggle to keep up with themselves and agents.
Too much for one designer to handle, but designers are still involved. Consider features/products in their adoption cycle: early preview vs. mature.
Launch as research preview with caveats about flaws. Quad Co-work (likely Claude Co-work) example: labeled research preview, worst it's ever going to be like models, but powerful benefits outweigh cons after internal testing. Commit to iterating based on feedback, show continuous shipping and improvements to maintain trust.
Losing trust happens if you release early and then do nothing. Possible to release early and maintain brand by responding and improving. Fun for designers: immediate feedback and learning.
Building trust through speed, making people feel heard, fixing based on feedback.
Anthropic labs excel: team tweets responses, ships fixes quickly (e.g., "we fixed this yesterday"), enabled by Claude coding fast.
AI progression: tab-completing code → 100% AI-written code → AI reviewing its own code → Claude helping decide what to build (Boris example).
AI will get better at taste, judgment, design, but humans still needed to decide what gets built and what matters. Hardest software work isn't building: disagreements on features. AI can weigh in with data but can't resolve human disputes.
Judgment on what to do next remains human (not just aesthetic taste). AI could provide data and rationale for decisions, but humans still accountable like engineers for Claude-generated code.
Re-evaluate assumptions: AI took over coding faster than expected (best engineers trust it without review). Likely to excel at PM/design judgment too.
Radiology analogy: humans sign off for liability, though not ideal job.
Combination of chatbots/terminals and tactile UIs with clicking. Claude added widgets for questions, weather, stocks—well-received for efficiency over typing.
Chatbot paradigm enables infinite flexibility not possible with baked UIs. Chat/terminals won't go away. UIs for specific tasks will be AI-generated, not hand-coded.
Innovations like OpenClaude, Claude MBOT via WhatsApp, Telegram, SMS: another chatbot form unlocked access. Talking scales across intelligence levels (200-300 IQ), handles growing model smarts.
[Sponsor: Omni for AI analytics with semantic layer, testing, permissions. Used by Perplexity, DBT, Buzzfeed. omni.co/lenny.]
Jenny joined Anthropic as IC, briefly managed, now full-time IC on Claude to stay close to work during rapid changes. Questioned if middle management persists.
Loves both IC and managing (reluctant manager). IC role built hard skills amid changing design processes, empathy for new ways of working, tool testing.
Previously Director of Design at Figma: max 12-15 designers + managers.
Managers valuable for teams: not pure people management (1:1s, career), but direction-setting + people management. Engage on work, create best environment.
Will return to management to help teams build best products ("whatever it takes"—direction or execution).
Advice: Design managers should do IC rotations like engineering (EMs rotate to understand tech). Design was more people-focused traditionally.
Rusty on IC return: crits (giving/receiving criticism)—vulnerable to share work and get regular feedback.
Leading design on Co-work. Internal prototypes/debates (Boris: make it fancy terminal). 10 days from internal to shippable (longer overall building).
Prototyped parts: to-do lists, multiple-choice, use case teaching on agent harnesses. Used what worked internally, shipped for signals, iterated.
Proudest: shipping it. Iterating homepage to feel like tasks for Claude/shared to-do list (randomizer for ideas, Claude's tasks needing attention). Explore form factor: always-on-screen or reaching other surfaces.
Not 10 days total: many explorations, people involved, idea iterated over time until right moment.
Co-work: "Claude with hands" (acts on computer), turns garbage into nice output.
One favorite use case for co-work is giving it a folder of stuff and extracting something good out of it regardless of contents.
Key traits for designers: resilience and willingness to adapt, try new methods and tools instead of sticking to old processes.
Three archetypes especially valuable:
- Strong generalists (block-shaped in T-shaped framework): 80th percentile in a few core skills, rare and hard to hire. Exciting because design roles are stretching into PM and engineering shapes; easy to flex and expand.
- Deep specialists (T-shaped with exceptionally deep tip): top 10% in industry, e.g., designers who are 50% software engineer (deep engineering expertise for working directly with models), or masters of visual design/icons to differentiate products when anyone can make anything.
- Crack new grads (early career, wise beyond years, humble, eager to learn): overlooked but valuable amid role changes. Blank slate, quick learners without baked-in processes/rituals; most companies hire seniors but new grads adapt fast.
Build a bunch of actual things, try stuff out, share with community. Don't feel limited by lack of experience—new grads are unburdened and see anything as possible.
Example from school: Socratica program (build and showcase like science projects). Examples: Claude robot project (assembling robots running on Claude); putting googly eyes on a bus in Boston. Builds agency and community of builders.
Helps to implement stuff (part of designer vocabulary now), but not necessarily code from scratch. As models/products improve, move up abstraction layers. Start using coding tools; be aware and know how to use them (e.g., not just learning React).
Claude not hirable yet—not strong generalist, deep specialist, or crack new grad. Good for first pass and presenting ideas, but nothing special. Good news for designers; curious if it can create amazing novel unique creative experiences.
Reject advice to avoid low-leverage tasks. Leaders you respect choose nitpicky/weeds tasks because they're doing it makes it high-leverage.
Examples:
- Senior leaders dogfooding product, repro-ing bugs, sharing logs with engineers: builds deep familiarity, shows care, rolls up sleeves.
- Senior leader fixing bugs (e.g., Mike Krieger putting in PRs): shows nothing below them, team vibe.
- Cultural: making personalized anniversary cards (vibe-coded nice). EA could do it, but leader doing it shows deep care.
Low-leverage becomes high-leverage via surprise/impact; makes leadership style stand out.
Encourages roasting not forced, but good sign of psychological safety: like roasting friends, shows comfort/pushing boundaries safely.
Team roasts leader (e.g., mocking her crit phrases like "Okay, what are next steps? How do we follow up?" by channeling her). Shows no fear, trust, personal knowledge of boundaries.
Balance: friends with reports? Create baseline psychological safety + high standards (not tension, complementary). Like tough parent: always there, won't fire on whim, wants best, high standards for great work.
Candid radical candor: care deeply + challenge directly.
From Evan Tana (partner at SPC/BC): 2x2 on founders/ideas as legible/illegible.
- Both legible: idea not novel, someone else will do it.
- Illegible idea interesting: on frontier, people don't get it yet or poorly told. VCs spot opportunities others miss.
Designers at frontier lab (Anthropic): spot illegible ideas internally, transform via storytelling/UX/form factor.
Example: Scanning Slack for energetic prototypes. Cloud Studio prototype (dense interface on agentic harness/Claude code): showed knowledge/skills/plans/to-dos/context/files. Initially illegible (didn't get it), but energy around it led to skills framework and markdown files instructing Claude. Pulled elements into Claude co-work.
Designers act like internal VCs on prototypes: dive into energetic but unclear ideas.
Patterns from early joiners at successes (Palantir, Stripe, Linear, Notion, OpenAI):
- Crazy illegible ideas (everyone laughs, impossible).
- People excited even if you don't get it.
- Top 1% founders (Anthropic has).
Humble, resilient amid changes. Span technical/prototypy to high craft/delivery. Hiring throughout year—excited about strong generalists/block and deep T specialists. Connect if un-scared.
Folks who feel like those archetypes resonate with them, but also folks excited about the technology who have been building a lot and want to be on the frontier.
Lightning round with five questions.
Two or three books recommended most:
- The Power Broker by Robert Caro (1100 pages). Worth reading end to end despite short attention spans. Spans someone's entire life (Robert Moses), shows how they change over decades. Nuanced view of a controversial figure. Reminder that careers are long and good for understanding how somebody gets things done.
- Insomniac City by Bill Hayes, partner of scientist Oliver Sacks around the time Sacks died. Beautiful, ethereal memoir of Sacks's last days and their love story. Makes you think about mortality, love, and life.
My goal is to create Renaissance humans.
Julie Zhuo, famed design leader, was reading The Power Broker recently—spreading in design circles.
A Sentimental Value (Norwegian film by director of The Worst Person in the World). Watched on a plane. Subtle pacing, writing, character relationships. Family drama about a house they've lived in their entire lives—the house is a character.
The Pit season 2. Everyone likes watching competent people do their jobs. Actors learn and memorize many terms; fast-paced with lots of movement in one shot.
Noah Wyle was in ER as a younger person and is now head of it.
Retro (used for two years since launch, discovering new benefits). Small community photo-sharing app: share photos only from each week (not all time). No social media elements—no counts, no ads. After two years, look back at each year (e.g., this week two years ago) to relive each week of life. Beautifully crafted app; designers love it.
"It is what it is." Sounds defeatist but brings levity amid uncontrollable world events and industry changes—helps move forward.
From a 10-day meditation retreat: confirms dad's wisdom—don't cling, don't try to change, just accept.
Introspection with local notes folder (using iA Writer): 1:1 notes, random thoughts, tiny memos, interview notes. Use Claude to analyze for insights and create things.
Practical example: For hiring, articulate what to look for in design craft (many struggle to articulate). Had Claude read all notes (interview notes, memos), then created a rubric for evaluating craft in interviews. Reveals implicit things about oneself.
Prompt: "Hey, I have a bunch of interview notes and notes related to design craft in this folder. Read it and help me craft a memo rubric for how I assess craft in interviews."
Twitter/X: @jenny_wen (best place, not much on LinkedIn).
Be helpful: Send product feedback on Claude or anything Claude-related to make it better.
Keep Lenny's Podcast in your library
Save the videos and channels worth coming back to, and find them again in one place.
