AWS Developers

Multi-Agent Patterns: Agents as Tools

Chapter 10 of the Building AI Agent Harnesses with Strands Agents course. This is where we cross from single-agent patterns into multi-agent orchestration, and we start with the most intuitive version: an agent that can call other agents the same way it calls any other tool. The parent agent reasons about the task, decides it needs a specialist, and invokes a child agent as a tool call. The child runs its own loop with its own tools and its own system prompt, returns a result, and the parent incorporates it and keeps going. From the parent's perspective, the child agent is just another tool with a name, a description, and a return value. This course teaches developers how to build AI agents and agent harnesses using the open-source Strands Agents SDK. The course starts by teaching what the difference between an agent and agent harness is, and progresses from fundamental concepts (the agent loop, model providers, tools) through intermediate patterns (hooks, plugins, steering, conversation management) to advanced multi-agent orchestration (agents-as-tools, graph workflows, agent swarms) and finally production operations (guardrails, observability, evaluation, cloud deployment). Reading: https://go.aws/4aLDDss Github Repo: https://go.aws/3SUiHtd Strands Agents SDK: https://go.aws/4bzWVBc Follow AWS Developers! 🆇 X: https://go.aws/4aHNuiX 💼 LinkedIn: https://go.aws/4vWfqYL #AWS #AgentHarness #StrandsAgents