Ryo Lu (Cursor): AI Turns Designers to Developers

Ryo explains how software development fragmented over 15 years into siloed roles (designers in Figma, PMs in docs, engineers in code), each with their own tools and artifacts. With Cursor, these boundaries are collapsing - designers can now build and ship real code in minutes instead of waiting months for designs to be implemented at 20% fidelity.

• •Traditional workflow: designers create mocks in Figma, wait a year, get 20% of what they wanted shipped
• •Cursor enables 60-70% accuracy on first shot, with rapid iteration replacing slow handoffs
• •AI agents can understand code, Figma mocks, Notion docs, and meeting notes - connecting all artifacts
• •The code base is the 'truth' - all roles (PM, designer, engineer) are ultimately modifying code
• •Cursor supports multiple workflows: tab completion for clean coders, agent mode for rapid prototyping

Discussion of how different roles (designers, PMs, engineers) have been siloed in their own tools and formats, leading to organizational friction. Cursor aims to unify these by making code the shared truth that AI agents can help everyone interact with, regardless of their technical background.

• •Each role developed habits around their own tools - designers in Figma/Sketch, PMs in Google Docs, creating inertia
• •Organizations 'ship their org chart' - fragmented teams create fragmented software
• •AI agents can synthesize past (accumulated knowledge), present (running code), and future (vision/planning)
• •One agent can serve different users with different views - code editor for engineers, visual interface for designers, documents for PMs
• •The goal is one base experience that works for anyone, with customization for specific needs

Ryo challenges the common discussion of 'taste' in design, arguing it's really about self-selecting boundaries of what is good based on what you've seen. LLMs have seen everything but lack opinion - humans must provide the constraints and preferences to avoid 'AI slop'.

• •Taste is ambiguous - it's really about selecting from options based on past exposure and making connections
• •LLMs have seen everything but are confused about preferences (e.g., thinking everyone wants purple gradients)
• •AI provides excellent baseline quickly, but humans must add opinion and boundaries on top
• •Cursor's 'plan mode' generates specs that users can refine - collaboration between AI capability and human direction
• •Long-running, non-specific prompts don't work - humans must specify what is good and right to avoid generic output

Ryo traces how early computing had generalists who designed everything from architecture to UI, then roles fragmented for optimization. Now AI tools like Cursor are enabling a return to individual builders who can handle the whole stack, with agents covering their weaknesses.

• •Early computing: researchers built entire systems - low-level architecture, UI, everything - as whole products
• •Modern era: roles fragmented into boxes optimized for cost, losing sight of the ideal whole
• •Ryo's personal journey: started building full stack, became 'product designer', stopped coding, watched designs die in meetings or ship as YouTube videos
• •With Cursor: designers can build, focus on their craft, let agents handle the rest
• •Strong teams can assemble specialists (infra, frontend, PM) who cover each other's weaknesses with shared tools and codebase

Ryo reframes design as designing concepts and systems at every layer - from core data models to UI - not just visual aesthetics. He uses Notion as an example of pure conceptual design where blocks, pages, and databases are simple JSON objects that enable infinite flexibility.

• •Design includes architectural decisions, ML infrastructure, CSS - not just aesthetics
• •Notion's core: blocks (JSON objects), pages (arrays of blocks), databases (pages with properties) - everything nests
• •Design is finding the best configuration and simplest state - fewest concepts to do the most for most people
• •Question isn't '6px or 4px border radius' but 'what's the minimal system with fewest code paths'
• •Beauty comes from putting all layers together coherently, not just the visual surface

While Cursor focuses on professional developers, non-technical people are already trying to use it. Ryo discusses plans to make it more welcoming without creating separate products - starting with simpler onboarding like showing an agent view instead of 'open project, clone repo, SSH' buttons.

• •Cursor intentionally made hard to access for non-technical users initially, but they're coming anyway
• •Current barrier: opening Cursor shows 'open project, connect to SSH, clone repo' - incomprehensible to beginners
• •Solution: start with blank agent view where people can just begin doing things
• •Designers could chat with browser preview, select elements visually, make changes that immediately happen
• •Not creating new products - same Cursor with different pre-configurations and packaging for different users

Ryo contrasts two philosophies: user-centric purpose-built tools (Webflow, v0, Asana) versus system-centric universal tools (Notion, Cursor). Purpose-built apps are 'selfish' - they silo workflows and can't grow without adding complexity. Universal tools with simple core concepts can serve everyone if designed right.

• •User-centric approach: identify problem, build specific solution - easier but limits growth from the start
• •Purpose-built apps create islands with proprietary formats, siloing people and workflows
• •Asana example: everything revolves around tasks and projects - naturally limits what it can do
• •Universal apps are harder to get started with but can be fixed with better packaging and AI assistance
• •Ryo's preference: build something that works better for everyone rather than optimizing for one group
• •Same AI agent architecture can take different forms - ChatGPT, Cursor, Replit, v0 are fundamentally similar

Ryo describes AI as a universal interface - at minimum a prompt and response - that can be shaped into different forms for different users. The key is designing mechanisms to transform the same underlying AI into optimal workflows for each person, not forcing everyone into a chat box.

• •Bare minimum AI interface: prompt input, get response - can be shaped into chat, sidebar, inline selection, or completely transformed
• •Just chat is a bad experience - blank input is intimidating, requires right questions, unclear what response you'll get
• •Need to design mechanisms to transform input/output into familiar workflows for different users
• •Examples: tab autocomplete for coders, visual element selection for designers, not walls of text
• •Underneath it's the same AI, agent, architecture - but each person finds the shape that fits them

Discussion of how constraints aid creativity and how Cursor balances simplicity with power user customization. Designers should focus on minimal abstractions and systems that handle permutations, not fixed button layouts. Users can customize everything from models to UI to keyboard vs mouse preferences.

• •Cognitive constraint: limit of concepts you can expose to any individual at any given time
• •Space constraint drives prioritization - what's most important doesn't change much
• •Build layers: 80% defaults for simplicity, then expose power user features and different archetypes
• •Cursor allows customization: custom agent modes, model preferences, tools, prompts, preview types, colors, keyboard/mouse
• •Designers think about minimal set of abstractions and systems to handle all permutations, not fixed layouts
• •Unlike CLI agents with tiny windows, Cursor lets you dig in and understand how things work

Ryo shares his creative process - sporadic, not routine-based, involving writing, walking, sketching, and coding. His RyoOS project recreates retro Mac interfaces (System 7, Mac OS X, Windows 95/XP) to show that core computing concepts haven't changed since 1984 - we're still using the same timeless patterns.

• •Process: writing in bullets, walking with phone/Notion, sketching, building prototypes - leaving blank space for ideas to simmer
• •Inspiration from everything: print design, films, music, art, nature, biological systems - not just software
• •RyoOS started as soundboard gift for Notion team, evolved into full retro OS with multiple themes
• •Built entirely with Cursor in 3-4 months - started with 'make it retro macOS-y', kept adding apps and windows
• •Message: same concepts from 1984 to now - windows, menus, icons, paint, text editors - timeless patterns that don't change
• •Technical constraints of each era shaped how concepts manifested, but we're still doing the same fundamental things