Traditional UIs are built around predictable data structures—someone designs the dashboard, the tables, the charts ahead of time. But AI outputs are inherently unpredictable, changing in structure and format with every interaction. This video explores why conventional interfaces break down when AI enters the picture and examines two fundamentally different approaches to solving the problem. The first, MCP Apps, ships full HTML/JS/CSS rendering code from the server to be displayed in sandboxed iFrames—powerful but inconsistent across tools. The second, AI-driven data rendering, flips the model: the UI maintains a library of predefined visualization components (diagrams, tables, cards, charts), and the LLM decides which ones to use, returning only structured data with type hints rather than code or Markdown.

Through a live demonstration using a Kubernetes cluster, the video shows the stark difference between an AI agent cramming complex architecture information into ASCII tables in a terminal versus the same data rendered as interactive diagrams, styled tables, and information cards in a purpose-built Web UI. The key insight is that we're moving toward a future where LLMs don't just answer questions—they decide how answers should be visualized and how users should interact with them. Industry efforts like Google's A2UI, OpenAI's Open-JSON-UI, and Vercel's json-render all point in the same direction: agents output declarative data, and clients render it with their own component libraries. The future of UI design may not be designed by us at all.

#AIUserInterface #MCPApps #DevOpsAI

Consider joining the channel: https://www.youtube.com/c/devopstoolkit/join

▬▬▬▬▬▬ 🔗 Additional Info 🔗 ▬▬▬▬▬▬
➡ Transcript and commands: https://devopstoolkit.live/ai/stop-designing-uis-for-ai-let-the-llm-decide-what-you-see
🔗 DevOps AI Toolkit: https://devopstoolkit.ai

▬▬▬▬▬▬ 💰 Sponsorships 💰 ▬▬▬▬▬▬
If you are interested in sponsoring this channel, please visit https://devopstoolkit.live/sponsor for more information. Alternatively, feel free to contact me over Twitter or LinkedIn (see below).

▬▬▬▬▬▬ 👋 Contact me 👋 ▬▬▬▬▬▬
➡ BlueSky: https://vfarcic.bsky.social
➡ LinkedIn: https://www.linkedin.com/in/viktorfarcic/

▬▬▬▬▬▬ 🚀 Other Channels 🚀 ▬▬▬▬▬▬
🎤 Podcast: https://www.devopsparadox.com/
💬 Live streams: https://www.youtube.com/c/DevOpsParadox

▬▬▬▬▬▬ ⏱ Timecodes ⏱ ▬▬▬▬▬▬
00:00 Web UIs for AI
01:10 Why Traditional UIs Fail AI
06:22 MCP Apps: Server-Side Rendering
09:01 AI-Driven Data Rendering
12:18 Future of AI User Interfaces

DevOps Toolkit's resource offers insights, tutorials, and resources for DevOps engineers and practitioners. Readers can learn about DevOps best practices, automation techniques, and tools for continuous integration and deployment. With articles, guides, and case studies, DevOps Toolkit provides  guidance and expertise for streamlining software delivery pipelines and improving collaboration between development and operations teams.

DevOps Toolkit

Traditional UIs are built around predictable data structures, but AI outputs are inherently unpredictable. Two approaches to solving dynamic AI visualization are explored: MCP apps (server ships HTML/JS rendering bundles to the client) and a component-library approach (LLM returns structured JSON with type hints, client renders using its own predefined components). The second approach is argued to be superior for specialized agents, keeping UI consistent while letting the LLM decide which visualization patterns to use. Industry players like Google, OpenAI, and Vercel are converging on declarative JSON output with client-side rendering. A reference implementation called DevOps-to is shared as a working demo.

Stop Designing UIs for AI - Let the LLM Decide What You See