Your UI Has Left the Building: Interfaces That Write Themselves
summary
Claim: LLMs generate interactive interfaces at runtime, not static screens, and three concrete protocols—AG-UI, A2UI, MCP Apps—cover different needs. Evidence: a 1950s pilot study (4,000 pilots, 140 measures, zero fit) motivates per-user UIs. Tools and mechanics: streaming lifecycle events, component catalogs tied to design systems, client-side rendering of incremental spec updates, and an MCP server that returns sandboxed UI resources to preserve host security. Working examples include streaming form fills and an agent-driven airplane seat picker rendered interactively in real time.
AI-written summary · about
chapters (10)
- 0:00Introduction to generative UI
- 1:12Problem with average user interfaces
- 2:40Challenges with agentic applications
- 3:02AGUI protocol overview
- 5:10ATUI protocol for UI generation
- 7:19Real-time UI streaming benefits
- 7:40MCP UI protocol explained
- 9:04When to use each protocol
- 10:10Combining protocols together
- 11:39Implementation recommendations
related talks
21mBeyond Components: The Future of AI UI
17mBeyond Components: Designing Generative UI for MCP Apps
28mLiquid User Interfaces for the Next Generation of AI Apps
30mEphemeral UIs That Learn, Adapt, and Vanish
33mIs Frontend Still a Thing? A UI Reset in the Age of AI Disruption
38mGenerative UIs: Using LLMs to Create Flutter Screens at Runtime
24mMind-Reading with Adaptive and Intelligent User Interfaces
15mThe Evolution of UI/UX Design: Exploring Latest Trends and the Impact of AI