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

Anthropic's design chief on why shipping beats the design process

AI broke the design process from the inside. Engineers can spin up several Claudes and ship faster than designers can produce mocks, so the research-and-diverge ritual designers were taught no longer fits the pace. Design is reorganizing into execution support and short-horizon vision work, and its remaining leverage is judgment, accountability, and shipping early enough to learn.

Takeaways

• The design process designers learned is dead. Her Berlin talk, Don't Trust the Design Process, called research, discovery and diverging and converging gospel. Within three or four months of Opus 4.6 and the wave of people discovering Claude Code, she says it felt outdated.
• Design work is splitting into two jobs. One kind supports implementation and execution, the other creates vision and direction. Vision now runs on a three-to-six-month horizon instead of a multi-year plan, and it is often a working prototype rather than a beautifully told deck.
• Her own time moved from mocks to code. Mocking and prototyping fell from 60-70% of her work a few years ago to 30-40%. A similar share goes to jamming and pairing with engineers, plus some implementation in code, sometimes by at-mentioning Claude in Slack on a bug and picking up the pull request.
• Trust comes from iterating after release. Claude Cowork shipped as a research preview with known flaws, forced out within 10 days, because the benefits outweighed the cons. Wen says releasing early hurts a brand only if nothing follows, so the team commits to responding to feedback.
• AI will improve at taste and judgment. Wen expects models to get better at taste and judgment, and says people may be holding too tightly to the idea that a human always knows best. Someone still has to decide what gets built and be accountable for it.
• Managers should take the low-leverage work. Wen was taught to categorize work and drop low-leverage tasks, but the leaders she respects most take those on themselves. She pairs high standards with a baseline of psychological safety and favors engineering-style IC rotations for design managers.

The legibility framework

• Evan Tana, a partner at SPC, drew a two-by-two of legible and illegible founders and ideas.
• A legible founder with a legible idea is probably not onto something novel.
• Wen used the framework to spot Claude Studio as an illegible prototype.
• Parts of that prototype fed into Claude Cowork and into the markdown files that instruct Claude.

What comes next for design tooling