I'm the creator of Manifesto AI.
I've noticed that while LLMs are getting smarter, their ability to interact with complex Web UIs is still fragile. Agents usually have to "guess" DOM selectors or rely on vision, which leads to hallucinations and broken workflows.
I realized that for AI to be useful in SaaS/B2B software, we don't need "Generative UI" (Text-to-App); we need a deterministic "State Layer" that agents can understand and control directly.
So I built Manifesto. It's a schema-first UI engine where:
1. You define the form/UI as a JSON Schema.
2. The engine renders it (React/Vue).
3. Crucially, it exports a "Semantic Snapshot" to the AI Agent.
Instead of parsing pixels, the Agent receives a clean JSON state (values, validation rules, available actions) and dispatches "Intents" (e.g., `setValue`, `submit`) to the engine.
Disclaimer: I built the core engine and this demo in just 4 days. It is currently in a very early Alpha (v0.1) stage.
I’m sharing this early because I want to validate if this "Intent-to-State" architecture makes sense to other developers.
I'd love to hear your feedback on the approach. Roast my code or the concept!