A2UI & Generative UI
AI-driven user interfaces that generate and adapt components based on intent.
Note: I can't directly show the prototypes I built, but I can share the references and research frameworks I used while exploring A2UI. This is early-stage research—not production-ready, but useful for understanding the space.
What is A2UI?
A2UI (Agent-to-UI) and Generative UI represent a paradigm shift in how interfaces are designed and built. Instead of pre-designing every screen, we specify intent and let AI generate appropriate interfaces. The key insight: defining what to explore is more important than building the prototype.
I used Claude Code to explore these concepts while understanding the boundary: this is research-grade exploration, not a production framework.
Exploring from Use Cases
Rather than building one "generative UI," I mapped the space by exploring different user roles and what they need (connecting to the "role" concept from prototype research). For each role, I designed interactions that would emerge:
Researcher Role
How might an AI generate interfaces optimized for exploration and hypothesis testing? What components matter when discovery is the goal?
Decision-Maker Role
What would an interface look like designed for quick judgment and trade-off evaluation? How does AI prioritize information density vs. clarity?
Executor Role
How does AI design for action and efficiency? What components reduce friction for users just trying to get something done?
Learner Role
How does generative UI explain itself? What if the interface adapted to teach as it generates, scaffolding understanding alongside functionality?
Interaction Patterns
For each role, I explored how interactions might differ when they're AI-generated:
- Fluidity — Can the interface morph as the AI's understanding of your goal evolves?
- Transparency — Does the AI show its reasoning for why this layout, this grouping, this hierarchy?
- Control — Where's the line between AI deciding and you deciding? How do you push back?
- Consistency — If every interaction generates a new UI, how does the user build mental models?
Research References
- A2UI Composer — Framework for thinking about agent-driven interfaces
- Generative UI research from Vercel and Anthropic on how to design for AI-generated components
- "What do prototypes prototype?" — Mapping this to the question: What does a generative UI prototype?
The Big Question
If interfaces can be generated, what's the role of design? The answer: design moves upstream. Instead of designing screens, you design the principles that guide generation. You decide: for this role, what matters most? Speed? Exploration? Control? Learning? The AI generates in service of that intention.
- How component generation is changing interface design
- The role of AI in scaling design systems
- Balancing automation with human control