Building Applications with AI Agents

Chapter 5. Orchestration

Now that your agent has a set of tools that can be used, it’s time to orchestrate them to solve real tasks. Orchestration involves more than just deciding which tools to call and when—it also requires constructing the right context for each model invocation to ensure effective, grounded actions. While simple tasks may only need a single tool and minimal context, more complex workflows demand careful planning, memory retrieval, and dynamic context assembly to perform each step accurately. In this chapter, we’ll cover orchestration strategies, context engineering, tool selection, execution, and planning topologies to build agents capable of handling realistic, multistep tasks efficiently and reliably. As we can see in Figure 5-1, orchestration is how the system utilizes the resources at its disposal to address the user query effectively.

Figure 5-1. Orchestration as the core logic that handles user queries and coordinates calls to the foundation models, external and local tools, and to various databases to retrieve additional information.

Agent Types

Before diving into specific orchestration strategies, it’s important to understand the different types of agents you can build. Each agent type embodies a distinct approach to reasoning, planning, and action, shaping how tasks are decomposed and executed. Some agents respond instantly with preprogrammed mappings, ...