← All speakers

Ryo Lu

Ryo Lu is a product designer, former head of design at Cursor, and creator of ryOS, an open-source desktop environment built with AI. His work gives people ways to reshape software: sharing content across Notion pages, bringing AI into existing editing workflows, and turning design ideas into working applications.

On this page

From interface experiments to Notion’s building blocks

Ryo’s interest in interfaces began after moving to Canada as a child. Fascinated by an Apple computer whose display seemed to float, he recreated its appearance on the computer he already owned, changing icons and fonts and learning to draw glossy buttons. Making something feel right came before having a professional reason to make it.

His career included product-design roles at Asana and Stripe before he joined Notion in 2019. Over roughly five years there, he worked as a designer and design lead on software that let people assemble their own documents, databases, and workflows.

Ryo helped design Synced Blocks, which let the same content appear in several places and update everywhere when edited once. The underlying idea came from Ted Nelson’s transclusion: including content while preserving its connection to the original. Ryo and the team made it approachable through copy and paste. They also had to show that an edit would affect more than the page in front of the user. Early designs created awkward spacing; another prototype moved text as users entered and left editing mode. The team settled on a lighter outline that appeared as the pointer approached and became more prominent during editing. Shared content could remain unobtrusive while reading, with its editable extent becoming visible when needed.

His work on Notion AI brought AI into those existing ways of working. An empty page called for drafting; existing content suggested continuation or summarization; selected text called for editing. Comparing an AI revision with the original gave the user a concrete choice before replacing anything. The team packaged common actions for immediate use while preserving custom prompts and reusable AI blocks for people building their own workflows.

Turning designs into working software

Ryo moved from Notion to Cursor in early 2025 and led design there until August 2026. Having spent years watching designs stall in meetings, he valued being able to implement and test an idea himself—a concern central to his December 2025 conversation with Jennifer Li.

His own ryOS makes that practice tangible. It brings a windowed desktop into the browser, with classic Macintosh and Windows themes, a file system, drawing and writing applications, music tools, and an AI assistant that can operate applications. The familiar desktop provides a place to experiment with AI alongside things people can open, move, edit, and keep.

In his writing on design, Ryo connects his Notion and Cursor work through decomposition and recombination. Notion’s blocks, databases, views, and relations support workflows their designers need not specify individually. Cursor extends that possibility to generating software from a person’s description. For Ryo, designing the underlying parts and relationships remains essential even as AI makes more of the implementation possible.

Leaving room to discover what you want

In his AI Engineer Shanghai talk, Ryo examined a difficulty that faster implementation does not resolve: a complete AI answer can settle choices before a person has considered them. His journal example makes the problem concrete. A request to record his day produced mood scores, writing streaks, reminders, and weekly reports, each with a plausible justification. Ryo would remove the streaks because missing a day would make him feel indebted. Someone trying to establish a habit might want to keep them. Trying a change helps the person discover what the journal is for.

That concern leads to specific interface choices:

  • Make proposed changes available to edit. Ryo contrasts a service counter that returns a finished version with a workbench where someone can inspect a plan, preserve a reminder slated for removal, and adjust the result in a live preview. The person can change one part without accepting or rejecting the whole answer.
  • Let people reclaim delegated work. Files, plans, edits, and differences between versions offer ways to understand and intervene. Ryo does not expect everyone to read every line of code or confirm every step. He wants a way into the work when someone needs to understand or change it.

Ryo distinguishes tasks he is happy to hand off, such as moving data or converting files, from work through which he is still forming an opinion. He also resists treating every gain in speed as a reason to produce more. He describes leaving San Francisco for Taipei partly as a way to stop explaining everything in his life through its usefulness to work. An empty calendar did not immediately bring certainty; it made the question of what he wanted to do harder to avoid.

Other people remain part of that discovery. A colleague who finds a detail confusing or insists on an arrangement Ryo would dismiss can reveal something he missed. AI’s patience and convenience can help, but always choosing the easiest respondent risks losing the experience of working through disagreement. Ryo wants tools that leave room for that exchange—and for making things together that none of the participants would have imagined alone.

1 conference talk

Loading more talks and highlights…