Generative UI Global Hackathon: Agentic Interfaces

Register only with this link : https://atlanta.aitinkerers.org/p/generative-ui-global-hackathon-agentic-interfaces-atlanta

# Generative UI Global Hackathon: Agentic Interfaces

A synchronized, heads-down build session happening across AI Tinkerers
cities worldwide - one global challenge, one submission platform, one day.
For three years, every “AI-powered” app has been a chatbot wearing a trench
coat. Text in, text out. On May 9, we take the training wheels off.
This is a build day for Generative UI - agents that don’t just return text,
they render complete, interactive interfaces on the fly. Forms, dashboards,
approval flows, entire applications - generated natively from agent output.
Presented by AI Tinkerers, Google DeepMind, and CopilotKit.

***

### Enter The UnHackathon

• No pitches
• No panels
• No chat-wrappers

Come build for six focused hours with engineers tackling the same frontier
problem at the same moment around the world.
Short global kickoff. Heads down. Show and tell at the end. Working code over
slides.

### Format

• 12:00 PM - Doors open (check-in, food, team-up)
• 12:30 PM - Global kickoff video: generative UI, the stack, the tracks
• 1:00 PM - Build. Teams form, starter kits and credits distributed
• 5:00 PM - Show and tell. 2-3 minute demos. Working code only.
• 6:00 PM - Submissions close. Projects submitted to the hackathon platform.

### The Challenge: Beyond the Chatbox

We are looking for functional prototypes that push the boundaries of how users interact with AI. Participants will compete to build systems that demonstrate:

• Dynamic Component Generation: UI that adapts its structure and state based on the underlying model’s reasoning.
• Agentic Feedback Loops: Interfaces that allow users to steer autonomous agents through interactive, real-time visual elements.
• Latency-Optimized Rendering: Utilizing techniques like KV cache optimization or local execution (via Gemma 4 or Muse Spark) to ensure fluid, zero-lag user experiences.
• Tool-Enabled Interfaces: UI that doesn’t just display data but provides the interactive hooks for agents to execute cross-app workflows.

Build with the protocols and frameworks defining generative UI in 2026:

• A2UI (Google) - Open protocol for agents to send interactive components. Apache 2.0.
• AG-UI (CopilotKit) - Protocol for connecting AI agents to frontend UIs.
• CopilotKit - Framework for embedding AI copilots into React apps.
• MCP Apps - Model Context Protocol applications rendered as interactive UI.

Food and drinks will be provided.

### Registration & Contribution

This is not for generative UI enthusiasts. This is for tinkerers.
We are screening for engineers who write code and will ship something by 6 PM.
Space is limited.
Not for everyone. For the ones who build.

## Global Sponsors

### Google DeepMind

Google DeepMind is the AI research lab behind Gemini and many of the open protocols defining how agents talk to interfaces, including A2UI - the open-source protocol letting agents send fully interactive UI components instead of plain text. Credits for participants distributed via the starter kit on event day.


### CopilotKit

CopilotKit is the open-source framework developers use to embed deeply integrated AI copilots into their React applications. It provides robust React components, an AG-UI transport layer, and an MCP-compatible client for building sophisticated generative UI experiences. The recent 1.5 release ships durable, long-running agents that feel native to your product. Starter templates, framework access, and full docs at docs.copilotkit.ai.


–

### Manufact

Manufact is the platform behind mcp-use, the fullstack open-source framework developers use to build MCP Apps for ChatGPT and Claude and MCP Servers for AI agents. The SDK ships in both TypeScript and Python with auto-discovery for interactive widgets, hot reload, and a built-in Inspector for testing. Manufact Cloud handles deployment, observability, and branch previews - push to GitHub and ship to production in under 60 seconds. Starter templates, the `create-mcp-use-app` scaffold, and full docs at github.com/mcp-use/mcp-use