The design process is dead. Here's what's replacing it. | Jenny Wen (head of design at Claude)

Stop clinging to the old design process: Don’t spend time doing full research or mockups for every step. Let engineers build the real thing (“let them cook”) and then step in to refine. AI tech moves too fast for long-term upfront planning.

Shift from mockup-maker to execution partner: About 30–40% of design work now is pairing or jamming with engineers in the trenches — helping decide and push features to ship (vibe coding together), not just handing off design files and walking away.

Use AI coding tools for last-mile polish yourself: Don’t wait for a ticket for an engineer to tweak CSS or small UI. Use AI tools in the IDE to adjust front-end code yourself so work moves faster.

Stop fake prototypes; test with the real AI model: AI models are non-deterministic. You can’t mock every state anymore. Build prototypes with real code so you can see how the system behaves with real data and prompts.

Shrink vision from 5 years to 3–6 months: Fancy long-term vision decks don’t cut it. Build working prototypes that point the team in the same direction in the short term.

Teach principles instead of owning every decision: Explain the “why” behind design and keep the design system clear so engineers can use AI to generate code or lay out UI correctly next time without waiting on you.

Build trust through speed: Ship features early as “Research Preview” even if they’re not perfect, and commit to listening and fixing fast. Shipping often and fixing quickly is the best way to earn user trust.

Keep Figma for planning and exploration: Code is the main deliverable, but Figma is still useful for quickly trying many options (e.g. typography or style) before investing in real code.

Be a lead who still does IC work: Whether you lead the team or own the system, spend time doing real hands-on work so you understand how AI tools are changing and empathize with the new workflow.

Do “low leverage” work for outsized impact: Testing heavily, hunting bugs, or opening PRs yourself builds culture and shows the team you care about details and lead by example.

Hire “block-shaped generalists”: When forming a team, look for people who are strong (e.g. ~80%) in several core areas (design, product, tech stack). They adapt best to agentic workflows.

Give “craft new grads” a shot: People without a rigid process baked in often learn and experiment with new tools faster. Look for side-project builders and people who aren’t afraid to try and fail.

Create psychological safety by letting the team “roast” each other: Environments where the team (including the lead) can tease each other show high safety. Once trust is there, you can raise the bar without burning people out.

Act as an internal VC (spot “illegible” ideas): Watch for odd ideas or prototypes in team Slack that seem unusable but people are excited about — then help structure the UX and turn them into shippable projects.

Use AI as an introspection tool: Dump raw notes, 1-on-1 takeaways, or scattered ideas into an AI workspace and ask it to extract frameworks, rubrics, or insights you can use to run the team or project.