MediAgent - LangGraph & LangSmith Implementation of the Microsoft Medical Agent

In this episode of The Build, Cameron Rohn and Tom Spencer walk through implementing the MediAgent using LangGraph and LangSmith to reproduce the Microsoft Medical Agent and evaluate practical deployment choices. They begin by outlining AI development and tools: instrumentation with LangSmith Execution Monitoring, cost controls like Claude Cost Cap, and onboarding flows via LangGraph Studio Onboarding tied into GitHub Version Control and Microsoft Teams Sync. The conversation then shifts to technical architecture decisions, comparing One-Shot Architecture Generation and Chain of Debate Orchestration patterns, tradeoffs for MCP tools, and hosting/edge concerns on Vercel with persistent state managed in Supabase. They explore building in public strategies and developer workflows, advocating a Feature Branch Workflow, public CI traces, and community-driven MCP tools contributions to accelerate trust and adoption. They also cover entrepreneurship insights, sketching product ideas such as an Automated Insurance Agent, AI Diagnostic Assistant, and a Workflow Performance Service monetized through instrumented metrics and enterprise integrations. Throughout, Cameron and Tom balance low-level API integration detail with higher-level startup tactics, examining how agents interact with APIs, telemetry, and developer ergonomics. They end with a forward-looking call for builders to instrument, iterate publicly, and prioritize observability as they ship next-generation AI products.