EP 23

In this episode of The Build, Cameron Rohn and Tom Spencer unpack memory systems and practical AI engineering for startup teams and makers. They begin by mapping the landscape of AI agent development and architecture, framing memory systems as a modular layer that supports retrieval, embeddings, and long-term agent state. They discuss concrete tools like Langsmith for orchestration, MCP tools for prompt and state management, and vector-backed patterns that integrate with Supabase for persistent storage. The conversation then shifts to developer tools and workflows, highlighting deployment choices with Vercel, end-to-end prototyping, and how various development approaches affect iteration speed. They explore building in public strategies next, sharing tactics for community feedback, open source collaboration, and monetization experiments that keep product-market fit transparent. They then address technical architecture decisions, weighing trade-offs in stateful vs. stateless services, consistency models, and observable pipelines that make debugging AI agents tractable. Interleaved are entrepreneurship insights on go-to-market, pricing, and sustaining community contributions. Smooth transitions tie these topics to practical examples and MCP tooling patterns that accelerate repeatable builds. They close with a forward-looking call for developers and founders to prioritize modular memory layers, rapid iteration, and public building to scale trustworthy AI products.