Ryo Lu, former founding designer at Cursor, uses a journal-app demo to examine judgment, creative identity, and collaboration in this 23-minute AI Engineer Shanghai talk.

A finished answer can hide unfinished decisions

  • Ask AI for a journal and it supplies mood scores, streaks, reminders, and weekly reports, each with a plausible rationale.
  • Those defaults quietly decide what recording a life should mean before the person using the journal has considered it.
  • Lu would remove streaks, reminders, and reports because missed entries feel like homework; another person might value that structure.
  • Choosing one feature to remove turns a vague preference into something concrete enough to discuss and change.

Build a workbench people can enter

  • A service-counter interface returns a finished artifact, leaving the user to accept, reject, or explain the request again.
  • A workbench exposes the proposed changes, supports intervention before execution, and offers a live preview for small edits.
  • Files, plans, and version differences provide places to inspect a decision and reclaim control.
  • Transparency means a usable route into the relevant detail; reading every line of generated code should remain optional.
  • Show important assumptions, keep changes small enough to try, and let people take back delegated work.
  • Preserve meaningful choices without requiring a confirmation button at every step.

Keep room for curiosity and other people

  • Lu traces his interest in design to recreating an Apple computer’s appearance through icons, fonts, and buttons as a child.
  • His aim is to retain that relationship with making things as AI makes output less scarce.
  • Hand off routine conversions freely, but leave room to write an initial thought yourself or put the tool down.
  • Daily journal pages and a freeform scrapbook reveal different needs behind the same request to record a life.
  • His jazz-band metaphor makes collaboration about listening, space, and finding the beat together, rather than playing the most notes.

Try one small change

  1. Pause at the next polished AI answer and identify a decision you have yet to think through.
  2. Pick a part you care about and change it.
  3. Show it to someone else and ask why they would choose differently.

Caveat: this is a reflective design talk with an illustrative journal demo. It supplies no measured comparison of interface effectiveness.

“Having an answer doesn’t mean I’ve finished thinking.” — Ryo Lu, 7:16