Agentic Architecture & Orchestration (D1)

Your support agent calls process_refund for a $240 charge. The model's response contains an assistant message with refund-confirmation text and a tool_use block invoking escalate_to_human, and stop_reason is 'tool_use'. Your loop must decide what to do before the customer sees anything. What is the correct continuation?

1. ATreat the confirmation text as the final reply, ignore the escalate_to_human block, and exit the loop since refund already succeeded.
2. BRe-invoke the model with a follow-up user message asking it to choose between confirming or escalating before acting.
3. CExecute escalate_to_human, append its tool_result, and re-invoke the model so it can produce a final end_turn message.
4. DStream the assistant text to the customer immediately, then execute escalate_to_human in parallel and continue the loop.