Justin Schroeder describes the challenges of building robust business agents, including integrations, observability, portability and fault recovery. He argues that tools and documentation alone do not solve these challenges.
He proposes composing a coordinator with specialist agentsA multi-agent system contains multiple AI agents that interact, coordinate, divide work, or influence one another while pursuing tasks. that have their own prompts, tools and limited context. Schroeder says this can improve token efficiencyAI token efficiency measures how effectively a model or workflow turns consumed input and output tokens into useful results., enable smaller models, restrict capabilities and allow parallel execution, while acknowledging that tasks must be defined carefully.
He outlines an agent architecture with tools, hooks, execution rulesAn AI agent harness is the software framework that packages a model with tools, instructions, context management, execution controls, and user interaction., sandboxed filesystems and code execution. He then describes recursively composed specialists for systems such as Salesforce and legal compliance.
Watch on YouTube
