Ep 19 - The week that was atlastic in AI. DGX Spark first thoughts demo, and Deepseek OCR oh my!

In this episode of The Build, Cameron Rohn and Tom Spencer discuss rapid developments in AI, demo the DGX Spark first thoughts, and react to Deepseek OCR's surprising accuracy while situating the news in practical build work. They begin by examining AI agent development and memory systems, unpacking architecture trade-offs for agent orchestration, stateful memory persistence, and observability using Langsmith and MCP tools to trace decision paths. The conversation then shifts to developer tools and workflows, with concrete examples of deploying prototypes on Vercel, persisting data with Supabase, and integrating Deepseek OCR into extraction pipelines. They explore building in public strategies next, including incremental releases, transparent metrics, and community-driven feedback loops that accelerate product-market fit. Chronologically, the hosts move into technical architecture decisions, debating vector store choices, microservices vs. monoliths, and cost-performance trade-offs when running on hardware like DGX Spark. Finally, they cover entrepreneurship insights—pricing, go-to-market experiments, open source contributions, and sustainable monetization informed by early user telemetry. They conclude by urging builders to iterate publicly with modular, observable architectures so developers and founders can learn faster and ship higher-quality AI products.