PRODUCTION SCENARIO
A team is building a purchasing agent with the Anthropic SDK. Every place_order call must be shown to a named human for sign-off before it executes, and every decision must be written to an audit log.
Which implementation should the team choose?
Answering here is anonymous. Nothing is saved unless you sign in.
Show answer and explanation
Answer: Write the tool-use loop by hand and dispatch each place_order call to a review queue
The tool runner exists to execute tools automatically, and the docs send you to the manual loop when you need human-in-the-loop approval, custom logging, or conditional execution. Registering place_order with the runner lets it fire before anyone sees it. tool_choice none stops the model calling the tool at all, so an operator retypes the order and no tool call is logged. Structured outputs move the order into JSON and leave the same unreviewed execution path in your own code.