Conversational Guidance
The user gives fuzzy, incomplete info. The agent reasons through it, calls the right tools, and asks targeted follow-up questions only for the details it can't figure out on its own.
The conversation
Open the chat UI side by side. Copy each prompt, send it, and watch the agent work.
- 1You
I want to return something. I'm Emily Chen, I think the order was 104-something from around April 14
- Agentsearch_orders finds Emily's ordersget_order matches "104-something, April 14" to ORD-10478 ($299)
Shows the item (ComfortEdge Office Chair) and asks: "What's the reason for the return?"
- 2You
Yeah the armrest broke after just a week
- Agentrequest_return processes the defective item return
Returns the return ID, refund amount ($299 to Amex), and a return label link.
What it looks like
ETC capabilities demonstrated
Ask clarifying questions
The agent knows request_return needs a reason. Rather than guessing, it asks the user for the one piece of information it can't figure out on its own.
Chain tools intelligently
"104-something from April 14" is enough. The agent calls search_orders, then get_order, compares dates and IDs, and deduces the right order without asking the user to be precise.
Follow instructions closely
The agent only asks for what it truly needs. It doesn't ask for the order number (it found it), the item (there's only one), or the email. Just the return reason, because the tool requires it.