Travel planning with Agentic AI
Samsung challenged us to design an agentic mobile assistant for a real-world travel scenario: helping a user named Sonia coordinate and book a weekend reunion with friends in Lucknow.
· High-fidelity screens for all touchpoints
· How the agent shows its reasoning, asks for consent, & gives the user more ways to control it
· One bonus: a live camera feature of our choice
Sonia says ‘look for flights for this’ over a friend’s message. Aris reads the screen to work out what ‘this’ means
Chain of thought made visible
Aris echoes her inputs back before acting, so she can correct a wrong assumption early
Consent gates the spend
Aris asks for the payment method, then needs an explicit confirm before it pays
Aris keeps Sonia in the chat
A floating window handles the questions (answered by tap, voice or keyboard), search & booking
Sonia asks for a cab to the airport with a stop for Ajay. Aris books it & drafts a message to him for her approval
The floating window stays with her on the home screen
Aris works across apps so Sonia never has to jump back and forth
Sonia stays in control
Aris drafts the message, but only sends it after she confirms
Sonia half-remembers a tip from Meera. Aris searches their past chats and pulls up the exact recommendations
TOLD ME ABOUT IN LUCKNOW?”
Edge case: human intervention
When Aris hits something only a human can do, like a captcha, it pauses and hands control to Sonia, then resumes when she hands it back
Sonia asks for a day plan around Meera’s picks. Aris drafts it, shows her the exact message, and posts to the group only after she approves
CHECK WITH THE GROUP IF THEY’RE UP FOR IT.”
While Sonia packs, Aris reminds her about Matthew’s hoodie, a promise from an old chat. A long press shows where it came from
Always showing its work
Aris links every suggestion straight back to the original chat, so recommendations feel useful instead of invasive
Sonia asks Aris to help empty the fridge. Aris reads what’s about to expire, suggests recipes, and orders what’s missing
Always showing its work
Payment stays off the shared fridge screen, Aris pushes the request to Sonia’s phone for a quick confirmation & payment, then takes back control
Sonia dictates her packing list. When Blinkit runs out of three items, Aris finds them on Zepto and shows both carts before she pays once
Aris explains store changes upfront
So she’s not caught of guard by two separate deliveries
Even with two different stores, she only has to confirm once
Sonia points her camera at the Bara Imambara. Aris highlights the exact structure on screen and explains what she’s looking at.
ViewCase
Adding a feature, or changing who can use it, meant asking a developer and waiting. The more features, the longer the wait.
ViewCase
Nobody tells them where to sit. They already know. Across 59 participants, kids turned out not to be rejecting each other at all. They were avoiding the teasing.
ViewCase
A game where bad money decisions make your tower wobble. Of 23 teens and 7 parents interviewed, the teens could tell a want from a need. Taxes sounded like something for later.