Lenny's PodcastJenny Wen, Claude head of design: Why design process is dead
Why non-deterministic AI states can't be cleanly pre-mocked anymore: Anthropic shipped Claude Cowork in 10 days by pairing with engineers, not mocks.
CHAPTERS
Why the classic “trust the process” design playbook is breaking
Jenny argues that the canonical diverge/converge, research-heavy design process designers were taught is no longer realistic. The speed of AI-enabled engineering has collapsed timelines and made fully polished mocks less central to shipping.
- •Traditional design process treated as “gospel” is effectively dead
- •AI accelerated engineering forces design to adapt (not the other way around)
- •Designers often can’t afford to produce “beautiful mocks” up front
- •The shift has accelerated even within months as tools improve
A new split: execution support vs. short-horizon vision setting
Design work is becoming stratified into two modes: helping teams execute quickly, and creating direction/vision. Vision work is still crucial, but it’s compressed to weeks or months and often expressed as prototypes instead of decks.
- •Mode 1: enable implementation—review, guide, connect, and polish as engineers ship
- •Mode 2: provide direction—align teams so outputs cohere into a product
- •Long-range visions (2–10 years) are less useful amid rapid model change
- •Modern “vision” is often a 3–6 month prototype that points the way
How widespread is the shift (and why there’s backlash)
Jenny sees the change spreading beyond AI labs: PMs and teams everywhere are spinning prototypes with tools like Claude Code and v0. But many designers push back because careers and identities were built around the old process and rituals.
- •Resonance across industry: teams feel they can’t do the old process anymore
- •PMs increasingly prototype directly with AI tools
- •Backlash: discovery and process steps feel foundational and threatened
- •Adoption varies by company maturity, culture, and comfort with change
Shipping-first design for nondeterministic AI products
Jenny believes faster execution and learning in production can produce better AI products, because AI behaviors are hard to predict in static mocks. For LLM features, you often need real models and real users to discover actual use cases.
- •AI products have too many states to mock exhaustively
- •Clickable prototypes can’t capture nondeterministic model behavior well
- •Real usage reveals emergent use cases you can’t anticipate
- •Speed + iteration becomes a core design advantage
A day in the life at Anthropic: keeping up is part of the job
At a frontier AI company, a significant part of Jenny’s day is understanding what’s happening across research, labs, and product teams. Internal prototypes, debates, and rapid iteration create a “constant catching up” dynamic that informs design decisions.
- •Tracking internal projects, prototypes, and code names to anticipate what’s coming
- •Staying close to model developments and philosophical direction debates
- •Curiosity as a professional advantage in a fast-moving environment
- •Design is increasingly about coordination and cohesion across parallel efforts
What designers actually do now: the new pie chart of time
Jenny describes a major rebalance: less time in mock/prototype creation and much more time pairing with engineers and implementing polish directly. Traditional research and prototyping still exist, but the toolset and proportions have changed.
- •Past: ~60–70% mocking/prototyping; now: ~30–40%
- •Now: ~30–40% “jamming” with engineers (consulting, feedback, pairing)
- •New slice: designers implementing and polishing in code
- •User research continues, but is complemented by faster real-world learning
Jenny’s AI stack and why Figma still matters
Jenny is all-in on the Claude ecosystem—Chat, Cowork, and Claude Code (often inside VS Code). She still relies on Figma for parallel exploration and micro-iterations that are harder to do in linear code-based workflows.
- •Primary tools: Claude Chat → shifting many use cases to Cowork; Claude Code in VS Code
- •Using Claude Code via Slack/mobile for quick fixes (e.g., icon tweaks)
- •Figma remains best for exploring 8–10 directions quickly
- •Figma excels at visual/interaction detail iteration without committing to a single coded path
Keeping quality and trust when everything ships faster
Jenny reframes quality management as a function of transparency and iteration. Shipping early is acceptable (especially as “research preview”) if users see rapid improvement, responsiveness, and follow-through—otherwise trust erodes.
- •Use staged expectations: early features framed as previews with known flaws
- •Trust comes from visible iteration and responsiveness to feedback
- •Brand damage happens when you ship early and then stagnate
- •“Build trust through speed” plus active engagement with users
Where humans stay valuable as AI gets better at taste and judgment
Jenny expects AI to improve meaningfully in design taste and judgment, challenging human exceptionalism. However, humans still need to decide what matters, resolve disagreements, and remain accountable for what ships.
- •AI will likely get better at taste/judgment than many assume
- •Hard part of product building is often deciding priorities—not implementation
- •AI can advise, but disputes and accountability still fall to humans
- •Analogy to engineering: AI writes code, but humans sign off and own outcomes
The future UI of AI: chat isn’t going away, but UI will expand
Jenny expects a hybrid future: conversational interfaces remain powerful for flexibility, while interactive UI elements accelerate common tasks. She anticipates more UIs being generated by models rather than hand-built each time.
- •Chat is a durable interface because it enables “infinite” flexibility
- •Widgets and interactive elements improve efficiency for specific tasks
- •A likely future: model-generated UI patterns for context-specific interactions
- •Terminals/chat coexist with more tactile, clickable surfaces
IC vs. management: why Jenny returned to hands-on work (and what managers should do)
Jenny moved from design leadership back to IC work to stay close to rapidly changing tools and workflows. She believes managers remain valuable, but the future manager must provide direction and understand the work deeply—not just do pure people management.
- •Returning to IC builds empathy and real-time understanding of new workflows
- •Middle management anxiety: questioning whether traditional roles persist
- •Future managers: direction-setting + environment-building, not only 1:1s
- •Analogy to engineering: managers often rotate through hands-on work first
Claude Cowork’s design evolution: many prototypes, then a fast push to ship
Cowork wasn’t invented in 10 days; it was refined over many explorations, then rapidly packaged for external release. The team shipped what was already working internally to get real signal, then committed to iterating in public.
- •Multiple internal prototypes explored form factors and interactions
- •10 days was the final stretch from internal state to shippable external product
- •Key interaction experiments: to-do lists, multiple choice prompts, teaching use cases
- •Primary win: shipping to learn, then iterating (e.g., homepage/task framing)
Hiring in the new era: resilience + three designer archetypes
Jenny prioritizes adaptability and tool openness, then looks for three profiles: strong “block-shaped” generalists, deep specialists, and exceptional early-career craft talent. She also advises new grads to build real things and find maker communities.
- •Resilience and willingness to adapt tools/processes is essential
- •Archetype 1: strong generalists (block-shaped, strong across multiple domains)
- •Archetype 2: deep specialists (top-tier in technical, visual, iconography, etc.)
- •Archetype 3: “craft new grad” (humble, fast learner, less process-burdened)
- •Advice: build and ship projects; community and output beats theory
Team culture and leadership: “low leverage” work, roasting, and legibility scouting
Jenny argues some “low leverage” manager behaviors are actually high leverage when done by leaders—like deep product dogfooding or thoughtful cultural gestures. She also views playful roasting as a signal of psychological safety, and uses a “legibility” lens to spot frontier ideas worth shaping into products.
- •Managers doing nitty-gritty product testing can be disproportionately impactful
- •Small cultural acts (cards, personal effort) can signal care and build teams
- •Roasting indicates comfort and trust when paired with high standards
- •Legibility framework: spot illegible ideas with energy and make them understandable via UX/storytelling
