EP - 7 - Building Agents with the A2A protocol and the ADK

In this episode of The Build, Cameron Rohn and Tom Spencer examine building agents with the A2A protocol and the ADK, focusing on practical architectures and developer workflows. They begin by unpacking AI agent development and tools, comparing the Grok 4 Model against evaluation datasets like the Amy Math Dataset and Humanity's Last Exam, and outlining Api integration patterns for memory systems and Comet AI Framework telemetry. The conversation then shifts to technical architecture decisions: Dual-Stack Tool Architecture, ARC General Intelligence Test considerations, and how MCP tools and Supabase-backed persistence interact with agent memory and state. They explore building in public strategies next, describing hands-on developer experiences hosted on Vercel, open demos of the Vending Bench Concept, and community-driven approaches like an AI Evaluation Community to validate agent behavior. Moving into entrepreneurship insights, they evaluate monetization for an Autonomous Vending Network and an AI-Operated Shop Concept, tradeoffs between hosted platforms and self-hosted stacks, and practical developer tooling that accelerates iteration. Throughout, they emphasize developer tools, CI patterns, and the role of Langsmith in observability. The episode closes with a forward-looking call: developers and founders should iterate publicly, instrument agents deeply, and build composable systems that balance rapid experimentation with robust architecture.