Members-Only
Recent Talks & Demos are for members only
You must be an AI Tinkerers active member to view these talks and demos.
How I Built, Broke, and Rebuilt a Multi-Agent Booking Flow From Complete Scratch in One Weekend
Learn how to build a multi-agent booking system, focusing on state management, data merging, and workflow progression challenges encountered when moving from a simple demo to a robust agentic workflow.
I built a live multi-agent intake and booking prototype that turns a basic chatbot into a structured agent workflow for qualifying a user, summarizing their situation, and moving them toward a demo booking flow.
The demo will focus on how the system is built, not on the product use case. I’ll show the live app, the /api/chat route, the agent workflow files, the structured JSON outputs, the state/lead-info merging logic, and the debugging logs that show how the agents move between stages.
The system uses multiple server-side agent roles:
Intake Agent
Qualification Agent
Summary Agent
Booking Agent
Each agent handles a different part of the workflow, but the hard part was not creating the agents. The hard part was keeping the app from forgetting what the user already said, merging extracted fields correctly, and deciding when the workflow should move from one stage to the next.
I’ll show the working system live, including the parts that broke while migrating from a simple manual router into an OpenAI Agents SDK workflow.