Case Study · 01 · 2022 — Present

LaylaAI

$1B+

In trips planned

Minutes

Intent → bookable itinerary

2,004,084

TRIPS PLANNED

00 / Origin

Before Layla, there was Adeo.

Adeo was the original prototype — a social, AI-assisted inspiration engine for travelers. We were exploring how people discover places, save ideas, and turn loose curiosity into a real trip. The early builds tested the hypothesis that became Layla: a single intelligent surface can replace the dozen tabs travel planning lives in today.

The three clips below are from that era — April 2023 walkthroughs of the inspiration feed, ecosystem concept, and the first end-to-end demo.

01 · Inspiration & Explore

Feed-based discovery — taste signals captured passively as travelers browse.

02 · Ecosystem Promo

The product vision: planning, booking, and social discovery in one surface.

03 · April Demo

First working end-to-end flow — the spine that became Layla AI.

01 / Context

Open-ended intent, bookable in minutes.

Travel planning lives in a hundred tabs — flights here, reviews there, a doc somewhere else. Layla replaces that with a single agent that learns each traveler's tastes, plans complete end-to-end itineraries, and books them.

The brief was deceptively simple: take an evolving LLM and shape it into a product that feels trustworthy enough to actually hand over your trip — and your card. In practice that meant designing for non-determinism, latency, partial knowledge, and a model that literally changes underneath you every few weeks.

We started narrow — a single conversational thread that could answer one question well — and widened the surface only when the agent's behavior earned it. Trust was the product.

02 / Capabilities

How the agent actually works.

Four surfaces sit behind a single conversational thread — search, planning, mapping, and inspiration — each tuned so the traveler never has to leave the chat to get something done.

Smart Search · Live Pricing

Real-time comparison across flights, hotels, trains, and car rentals. A flight prediction engine forecasts price trends; smart hotel search filters by natural prompts like “sea view with a rooftop pool.” Activities, experiences, and private transfers all bookable in one conversational flow — available in 16 languages.

AI Planning & Personalisation

Travelers share dates, interests, and budget; Layla builds a custom day-by-day itinerary in seconds. The conversational assistant learns preferences — romantic dinners, kid-friendly stops, off-beat adventures — and re-plans on the fly. Itineraries export as downloadable PDFs for offline use.

Visual, Interactive Maps

An Interactive Video Map overlays creators' travel videos onto destinations — Rome's Colosseum through a vlogger's lens, hidden bars in Tokyo. A Multi-Destination Route Map powers European rail journeys and USA road trips, while the Road Trip Planner suggests scenic stops, drive times, and overnight stays.

Media-Driven Inspiration

Viral travel videos and authentic creator content — drone footage of alpine lakes, foodie tours in Paris, street scenes in Bangkok — are linked directly to planning tools. Fall in love with a destination, add it to your itinerary in one tap.

03 / Who it's for

One agent, seven kinds of traveler.

The same underlying model adapts its tone, pacing, and recommendations to the trip in front of it — from a family half-term to a luxury honeymoon.

Families

Balance sightseeing with downtime; kid-friendly hotels, activities, and transport.

Couples

Boutique hotels, wine tastings, sunset cruises — romantic itineraries packed with memorable moments.

Solo travellers

Stay safe and flexible with curated neighbourhoods and local experiences.

Groups & friends

Coordinate schedules, share itineraries, and manage budgets together.

Road trippers & rail lovers

Scenic routes optimised for driving and train times so the journey itself is the trip.

Business & bleisure

Efficient schedules that mix business with leisure, perfect for extending work trips into vacations.

Luxury seekers

First-class flights, five-star resorts, private transfers, and exclusive experiences.

04 / Approach

Design the agent's behavior, not just the screens.

Taste discovery

A lightweight onboarding that captures preference signals the model can act on — without feeling like a form.

Intent → itinerary

A conversational surface that turns vague prompts into structured, editable plans.

Trustworthy booking

Designed the moments where the agent commits real money — with the receipts, transparency, and control travelers expect.

Adaptive behavior

Re-plans when weather, prices, or your mind changes. Surfaces what shifted and why.

05 / Principles

Heuristics for a moving model.

Show your work

When the agent commits to something — a flight, a hotel, a route — the surface should make the why visible without forcing the user to ask.

One question at a time

LLMs love long answers. People don't. Each turn should ask for one decision, then move on.

Editable, not magical

Every output is a draft. Swap, remove, re-plan. The agent proposes; the traveler still owns the trip.

Adapt out loud

When conditions change, surface the diff. Silent swaps erode trust faster than any bad recommendation.

Design the moment of payment

The booking screen carries the entire product. Receipts, fine print, and clarity matter more than novelty here.

06 / Outcomes

What it shipped to.

$1B+ in trips planned through the agent.
2M+ trips planned with an average 4.9★ traveler rating.
Live integrations with Booking.com, Skyscanner, and GetYourGuide — real inventory, real prices, one checkout.
A consumer-grade AI product used end-to-end — discovery, planning, booking, and in-trip support across 16 languages.
A design language and set of interaction patterns other agentic products are starting to look like.

Reflection

“Designing for an LLM is less about screens and more about behavior — what the agent does when no one is looking, and how it tells you about it after.”