Beyond the Chatbot: Gen UI & the Execution Trust Problem
summary
Demonstrates running LLM-generated TypeScript functions (not JSX) in Cloudflare dynamic worker instances backed by V8 isolates to execute in milliseconds while containing blast radius. Model gets a constrained tool that fetches predefined data (no raw SQL), execution runs with outbound policy set to global outbound null or an allowlist, and responses pass schema validation then map to trusted React components. Includes runtime CPU/time limits and ephemeral sandboxes; treats generated code as untrusted and validates both server and client.
AI-written summary · about
chapters (12)
- 0:00Welcome
- 1:52Key questions on Gen UI
- 3:39Chat blurb to Gen UI demo
- 6:28Live Gen UI demo
- 9:00How Gen UI code works
- 10:26Safe code execution
- 11:54Dynamic workers for security
- 12:39Preventing data exfiltration
- 14:04Validating output
- 15:27Security checklist
- 16:52Untrusted generated code
- 17:34Closing
related talks
20mCode Mode: Let the Code do the Talking
36mEvery API is a Tool for Agents with Code Mode
06mQ&A with Jòan Varvenne
38m🔥 Generative UI: The Future of Frontend (And Why You Should Care)
29mCoding Chat Bots
26mAutomate Your Life: Building a Workflow Engine in TypeScript
49mCoding Tomorrow
30mUnleashing AI in Your Web Apps