Descriptions:
Tyler Slaton, founding engineer at CopilotKit, delivers a technical talk at AI Dev 25 in New York City covering the AG-UI (Agent User Interaction) protocol—an open, event-based standard his team created for connecting agentic backends to frontend UIs. With CopilotKit now counting 30,000 GitHub stars, 200,000 weekly installs, and usage among 10% of the Fortune 500, Slaton explains why existing request-response paradigms break down for long-running AI agents and what AG-UI does differently.
The talk maps the current agentic protocol landscape: MCP (Model Context Protocol) handles context flowing into an LLM, A2A handles agent-to-agent communication, and AG-UI handles the client-server layer between agents and user interfaces. Slaton details CopilotKit’s role as a React consumer of AG-UI streams, enabling any AG-UI-compliant backend to power rich, interactive frontends without tight coupling.
The second half of the talk walks through distinct generative UI patterns—structured (component-mapped tool calls), headless (programmatic agent cloning with time-travel support), and open-ended (iframes and raw HTML)—weighing the tradeoffs of each in terms of design flexibility, backend-frontend coupling, and predictability. Developers building SaaS co-pilots or productivity tools on top of LLM agents will find this a useful framework for thinking about where AG-UI fits alongside other emerging agentic standards.
📺 Source: DeepLearningAI · Published December 05, 2025
🏷️ Format: Deep Dive
