Good agents don't just answer questions — they guide conversations. You'll design a full conversational script for a travel booking agent: mapping user intents, writing clarification flows, presenting options clearly, and refining prompts from real user feedback.
Voyager is a conversational travel booking agent. It books flights, hotels, and packages through natural conversation. Your job is to design the dialogue that makes Voyager feel like a brilliant travel consultant — not a search form.
Voyager users arrive with wildly different levels of specificity. Some know exactly what they want: "Business class, JFK to Heathrow, Oct 14, returning Oct 21, under $4,000." Most don't: "I want to go somewhere warm in November."
The agent's dialogue must handle the full spectrum — from decisive to undecided. It must ask exactly the right clarifying questions (not too many, not too few), present options in a scannable format the user can act on, recover gracefully from confusion, and feel like a conversation rather than a form.
The number-one failure mode of travel agents is asking too many questions before giving any value. Users abandon when they feel interrogated. The design challenge is gathering necessary information while always moving the conversation forward.
Every user message carries an intent — often an ambiguous one. Before writing dialogue, you need to know what users are actually trying to say, and the minimum clarification needed to move forward. Open all 5 intents, then answer the comprehension question.
Work through 5 critical dialogue nodes — choosing the right response strategy at each decision point. Your choices assemble into a complete, readable conversation script at the end.
A dialogue that works for one user type will fail for another. Test Voyager against three personas with different communication styles. Watch where it excels and where it breaks — those failures are your design inputs for the Refine section.
Real user testing surfaces friction points. Below are four actual feedback signals from the Voyager beta. For each: diagnose what went wrong in the original design, then see the corrected version.
You've mapped intents, written a script, tested personas, and refined prompts. Now assemble the final design canvas — the document a real product team would use to implement Voyager's conversational logic.
Drawn from every section. 70% to pass.