Food Ordering Agent
Designing the flagship voice-based ordering experience for Google Cloud and Papa Johns
Overview
Google’s Food Ordering Agent
As the UX Design Lead for Google’s Food Ordering Agent, I owned the design of a conversational ordering experience for quick service restaurants. Moving beyond the limitations of traditional chatbots, Food Ordering Agent is a purpose-built voice AI agent engineered for the high-volume, high-complexity realities of the food industry. Working with engineering, product, and research teams, I translated Google’s core voice technology into a multimodal customer experience adaptable across mobile, in-car systems, kiosks, and drive-thrus.
I worked most closely with Papa Johns, Google Cloud’s first retail partner to deploy their mobile food ordering agent. This experience was launched publicly as Lou AI.
Challenge
From proof of concept to nationwide launch
Our challenge was to take an early conversational-ordering prototype and turn it into a production mobile experience for one of the largest pizza brands in the world (serving 150M+ customers across 6,000+ restaurants) while building a white-label foundation that could flex to other quick-service brands and surfaces.
UX Explorations
How much voice do users want in a touchscreen experience?
How much ‘voice’ do people actually want? We explored three different ways to order, from a subtle assistant that helps out on a familiar screen to a fully hands-free experience where the conversation leads the way.
Hybrid Overlay
Voice as a secondary layer over traditional touch navigation.
Conversation-Led
A text thread that transcribes what the user says and the agent replies. Supplemented with contextual visuals from the menu. The full menu and cart are accessible via a bottom sheet.
Voice-Native
A “hands-free” approach that emphasizes a minimal UI to encourage natural conversation between the user and agent. The screen displays only the most recent agent response and the items in the user’s cart.
UI Evolution
Bringing visual affordances to a voice-first experience
Over a seven week period, we rapidly iterated on the design, prototyping each design to test how the human-agent conversation unfolded over the course of an order.
Customer Solution
Anchoring the experience on core customer needs and business value
While users showed a clear preference for an immersive, voice-native experience over a hybrid approach, research revealed a simultaneous need to verify order accuracy and access the full menu. To reconcile these competing desires, I designed a conversation-led interface that grants primacy to the real-time agent transcript while contextually highlighting relevant menu items within the dialogue.
This streamlined UI maintains essential on-screen utility—such as location management and exit controls—without cluttering the voice-first experience. To ensure the agent interaction remains the focal point, the comprehensive menu and persistent cart are tucked into a draggable bottom sheet, allowing users to dip into traditional navigation without breaking the conversational flow.
Enterprise Console
Solving the “black box” of agent management
From the early pilot stages, we heard that businesses felt that AI is a “black box.” We designed an enterprise console to give visibility and control to enterprise customers.
Understanding Performance
Provide a glanceable overview of agent performance through metrics like agent containment, task completion, crew intervention, average check size, and upsell revenue
Configuring Behavior
An “Upsell Engine” where brands set global guardrails and specific strategies (e.g., “Margin Maximizer” or “Inventory Optimizer”) and deploy them during specific day parts.
System Monitoring
Test the agent through a web simulator and track order history with real-time auditing.
Impact
Impact in the first months: higher conversion and order speed
“Compared with non-AI assisted orders, customers using Lou AI are converting at an 18% higher rate and completing their orders approximately three minutes faster.”
— Todd Penegor, Q2 2026 Earnings Call (source)
Google’s Food Ordering Agent was unveiled with Papa John’s at NRF in January 2026, and shipped publicly in the Papa John’s app as “Lou AI” in April 2026–serving over 150M+ customers across 6,000+ restaurants. Since launch, CEO Todd Penegor publicly cited Lou AI as a driver of higher order conversion and faster checkout.
Takeaways
Designing for emerging needs
Innovation at the Interaction Level
Naturally, there’s a lot of attention on AI products and AI experiences. But there’s another layer of improvement to be made on a much smaller scale, on the interaction level. Little user interactions that can speed up or improve existing interactions, whether that be selecting a menu item, filtering a list, etc. These improvements are more subtle, and less defined, but they add up quickly to big experiential improvements.
Design with Skepticism
We’re in a moment of terrific optimism and dynamism about the possibilities of AI. Designers should be a part of dreaming up what that optimistic future for users can be, but they are also responsible for not letting AI-mania get in the way of user experience. Identify unnecessary or unhelpful AI experiences which degrade user experiences.
Build for Learning
In the traditional design process, we start with a concrete user need or pain point and build a way to address it. In AI-first design, we recognize that needs are still emerging and unpredictable. Instead of a static product, we build a learning engine that users can tune and evolve over time. We aren’t just designing a solution, we are designing a capability that grows through every interaction.