Building agents: workflows and tool calls
Anthropic explains when agent autonomy helps and why clear tools matter.
Choose the architecture
The companion essay distinguishes predetermined workflows from agents that select their own steps. It presents prompt chaining, routing and evaluation loops, and recommends adding complexity only when simpler approaches fall short. The February 2025 discussion belongs to that earlier tooling period.
A first implementation exercise
For a follow-up exercise, use the separate tool-use guide. With API access and basic programming knowledge, define a harmless lookup with a clear input schema. Inspect the returned tool request, run the lookup in your application, and return its result. Compare the arguments with the original request before adding another operation. This exercise focuses on the boundary between a model selecting an action and your code executing it.
