A single-agent system uses one autonomous agent to perform the tasks needed to solve a problem or achieve goals. It is usually the smallest practical starting point: one agent receives instructions, selects from its available tools, and works toward the requested outcome. section 4.0 section 4.2.1
For simple use cases, one agent can be effective. As functionality grows, however, the agent may face too many tools, overly long instructions, overlapping tool descriptions, or difficult planning and focus-switching demands. At that point, splitting the system into specialized agents can improve reliability, cost, latency, and context management. section 4.0 section 4.2.1
From one research agent to a flow
A single research assistant can combine research tools, sequential thinking, and filesystem access in one context to find information, plan research, and save a result. This works initially, but extensibility breaks down as the agent must understand every tool. A multi-agent flow can instead assign research, planning, and filesystem work to distinct role-based agents. section 4.2.1
Use a single agent when the task is focused and its tools and instructions remain manageable. Consider an agent flow when the agent repeatedly chooses the wrong tool, carries irrelevant context, or must handle several distinct responsibilities. section 4.2.1 section 1.4
