Summary of "The design process is dead. Here’s what’s replacing it. | Jenny Wen (head of design at Claude)"
Key thesis
- The traditional, slow “design process” (long discovery, polished mocks, multi-year visions) is no longer the default.
- Rapid engineering iteration and AI tooling split design into two main modes:
- Hands-on execution support — pairing with engineers, implementation, last-mile polish.
- Short-horizon vision-setting — 3–6 month prototypes that point teams in a direction.
- Designers should stop acting as gatekeepers and instead enable fast iteration, keep product coherence, and decide what actually gets built and why.
Tools and AI stack used by designers (practical examples)
- Claude (chat) — primary conversational assistant.
- Claude Co‑work — long-running, agentic workspace described as “Claude with hands”; supports tasks, shared to-do lists, folder ingestion, and multi-step flows.
- Claude Code — code-generation and agents inside IDEs (commonly used in VS Code); useful for prototype iteration, PRs, and quick fixes.
- VS Code, Slack, mobile integrations — combined with Claude Code for quick frontend tweaks and deployment.
- Figma — essential for exploratory ideation, multiple-direction visual exploration, and micro-interaction design; complements code-driven work.
- Implicitly used tooling: research studies, surveys, design systems, and prototyping infrastructure.
Product features and UX patterns (Claude Co‑work & related)
- Agentic workspace: Co‑work ingests files/folders, generates plans/todos, and executes multi-step tasks with the user.
- Homepage evolution: surface tasks Claude is working on and present a shared to-do list between user and agent.
- Interactive widgets inside chat: richer UI elements such as multiple-choice prompts, preview outputs — a hybrid between chat and clickable UI.
- Rapid release pattern: ship early as a “research preview,” respond actively to feedback, and iterate quickly to build trust.
- Practical flows: give a folder of mixed content and have Co‑work extract outputs (synthesis, memos, rubrics); use Co‑work to analyze personal notes and produce hiring/evaluation rubrics.
Design → engineering reality & implications
- AI models are non-deterministic; many product states cannot be fully mocked. Real-model prototypes and shipping to users are necessary to understand use cases and limitations.
- Engineers can spin up experiments and ship fast (including multiple agent-backed flows); designers’ role shifts toward enabling coherent, high-quality experiments.
- Implementation-level design work (editing CSS, swapping class names, making PR polish) has become common for designers — pairing directly in code shortens the feedback loop.
- Chat/terminal interfaces provide a flexible medium to interact with LLMs; they will remain relevant, while richer generated UIs/widgets and surface-specific form factors grow in use.
Hiring, skills, and career advice
- Three designer archetypes to hire:
- Strong generalists (block-shaped): multiple ~80th-percentile skills across design/PM/engineering for flexible execution.
- Deep specialists (deep-T): top-10% experts in a niche (visual craft, iconography, deep frontend skills).
- Craft new grads: early-career, curious learners without entrenched rituals — high ROI in fast-changing environments.
- Managers should rotate back into IC (individual contributor) work periodically to stay current with tooling and to empathize with changing design work.
- Practical advice for new grads: build real things, ship prototypes, join communities, and share work — real artifacts beat theoretical portfolios.
- For senior designers: basic coding/tool familiarity (IDE, CLIs, code-gen agents) is increasingly useful even if deep engineering is not required.
Management and team practices
- Senior leaders doing “low-leverage” hands-on tasks (dogfooding, reproducing bugs, adding PRs, personal touches) can be very high leverage — these actions communicate care and build trust.
- Psychological safety: encourage candid feedback (even friendly “roasting”) to build trust and maintain high standards; balance warmth with directness.
- Designers in management should provide work-level direction (not only people ops); managers who can give concrete direction and participate in the craft are more effective.
Frameworks & internal product discovery
- Legibility framework:
- Founder-legible vs idea-legible/illegible.
- Designers at frontier labs should spot “illegible” ideas (internal prototypes with energy but unclear form) and translate them into understandable products via UX, storytelling, and form decisions.
- Example pipeline: an internal dense “Cloud Studio” prototype (illegible) → decompose into skills/markdown instruction formats → inform Claude Co‑work’s form and features.
Product process examples
- Co‑work’s launch involved many internal prototypes and experiments before the public 10‑day push; releasing early enabled fast feedback and iteration.
- Labeling early releases as “research preview” and committing to rapid iteration preserves brand trust when the team visibly addresses user feedback.
Limitations & where humans still matter
- LLMs are strong at first-pass ideation and surfacing options but are not replacements for full designers: they lack consistent taste, deep craft specialization, and accountable judgment.
- Humans remain responsible for high-stakes decisions, accountability, and dispute resolution about product scope and priorities, even as AI augments ideation and engineering.
Practical, shareable tips & examples
- Automate small frontend fixes via Slack mentions + Claude Code + PR workflow.
- Use Co‑work to ingest notes and produce hiring rubrics or syntheses — a real use case for introspection and hiring prep.
- Keep using Figma for divergent exploration and pixel-level polishing; use code/agents for convergent iteration and real-data prototypes.
Notable product comparisons and metaphors
- Jenny’s short descriptor for Co‑work: “Claude with hands” — an agent that can manipulate and synthesize your files and outputs.
- Chat remains a versatile interface because it scales across levels of intelligence and complexity; richer generated UIs will be produced by models and coexist with chat.
Speakers / sources
- Jenny Wen — Head of Design for Claude / lead design for Claude Co‑work; former Director of Design at Figma; ex-Dropbox, Square, Shopify.
- Lenny Rachitsky — Podcast host.
- Other referenced people/ideas: Boris (Claude Code), Evan Tan (legibility framework origin), Kevin (on talking as UI), Anthropic (company / Claude products).
Possible follow-ups (examples of extractable outputs)
- Extract practical prompts/recipes Jenny described (e.g., folder ingestion → rubric generation with Co‑work).
- Produce a hiring checklist based on the three archetypes.
- Summarize Claude Co‑work’s product features into a concise product spec.
Category
Technology
Share this summary
Is the summary off?
If you think the summary is inaccurate, you can reprocess it with the latest model.
Preparing reprocess...