Saw a talk from a dev at Amazon along these lines recently.
The general concept is called “server-driven UI” (SDUI) and they talked about experimenting with a completely AI/LLM-powered frontend. It has too many problems today for practical use (LLM FE sucks with accessibility, not to mention the overall cost!) so they instead tried a half measure.
Their FE team makes a series of generic components (“primitives”) and the AI then picks among them to “build” the FE on demand. That’s the “control the experience” thing you’re getting at.
They then (hand wave) allowed the LLM access to a customer data DB.
This unused experiment would let customers search things like “what movies will I like?” and get a cogent FE despite no engineer shipping that specifically.
The general concept is called “server-driven UI” (SDUI) and they talked about experimenting with a completely AI/LLM-powered frontend. It has too many problems today for practical use (LLM FE sucks with accessibility, not to mention the overall cost!) so they instead tried a half measure.
Their FE team makes a series of generic components (“primitives”) and the AI then picks among them to “build” the FE on demand. That’s the “control the experience” thing you’re getting at.
They then (hand wave) allowed the LLM access to a customer data DB.
This unused experiment would let customers search things like “what movies will I like?” and get a cogent FE despite no engineer shipping that specifically.