EP - 9 - AI Exec Orders, Qwen 3 Coder, JSON Veo3 Demo and Graph RAG Deep Dive with Neo4J

In this episode of The Build, Cameron Rohn and Tom Spencer dig into AI infrastructure, demo workflows, and pragmatic startup decisions for building machine intelligence products. They begin by walking through hands-on experiments with JSON Veo3 and the JSON Veo3 Demo running against a Cloud Hosted MCP Server, showing how MCP tools orchestrate AI agents and memory systems while integrating Langsmith for tracing and observability. The conversation then shifts to developer tooling and deployment patterns, comparing Vercel and Supabase for hosting frontends and vector-backed services, and highlighting Qwen 3 Coder as a case study for model selection and latency trade-offs. They explore technical architecture decisions around Graph RAG with Neo4J, selective transcript embedding, and composable AI workflows that combine retrieval, short-term memory, and agent planning. Along the way they unpack building in public strategies, from iterating transparently to using Langsmith Integration metrics to attract contributors, and they surface entrepreneurship insights including Hosted MCP as SaaS, a Vault Customer Success App, and a Security Compliance Service as potential monetization paths. The episode ties technical choices to business outcomes, leaving developers and founders with a forward-looking mandate to prototype quickly, instrument thoroughly, and build in public with clear product hooks.