Generating UIs by having LLMs emit raw HTML/JS? That's a risk pile, says Googler
rseroter · x · 2026-08-19
Google's Rich Seroter warns that "generative UIs" built by letting an LLM produce arbitrary HTML and JS code carry all kinds of risks. The post he shares doesn't say it outright, but makes the case that structured UI intent plus a component registry—the A2UI approach—is the better path, keeping generated interfaces within a controlled component system instead of executing free-form code.
More from coding & agent
- Puck now supports real-time voice for agent coordination — HankYeomans · 2026-08-19
- Code.Storage Opens Signups: Unlimited Git Infrastructure for AI Agents — dsp_ · 2026-08-19
- Using Directed Graphs for Predictable Agent Workflows — rseroter · 2026-08-19
- Dev vibe-codes a mobile shell over Maestro, prompts local agents from iPhone via Tailscale — cocktailpeanut · 2026-08-19
- LlamaParse Outperforms General VLMs in Document Visual Grounding — llama_index · 2026-08-19
- ComfyUI releases official local MCP open-source — crystal_alpine · 2026-08-19