Driverless taxi hits a new milestone: How Waymo's Gemini in-car AI assistant is transforming the riding experience

Waymo’s advancements in autonomous taxis are accelerating once again. According to renowned tech researcher Jane Manchun Wong’s latest discovery, Waymo is developing a comprehensive AI assistant system that uses Google’s Gemini technology to provide in-vehicle services for passengers. This finding reveals how self-driving car companies are integrating conversational AI into their operations, marking the evolution of autonomous taxis from mere “transportation tools” to “intelligent mobility platforms.”

In her blog investigation, Wong noted that she found a complete system prompt file within the Waymo mobile app code, internally called “Waymo Ride Assistant Meta-Prompt.” This file exceeds 1,200 lines and details the behavior standards for the AI assistant inside the vehicle. Although this feature has not yet been released in public versions, the complexity of the system prompts indicates that Waymo has high expectations for this in-car assistant.

Over 1,200 Lines of System Prompts: Gemini’s Role in Autonomous Taxis

According to the system prompts, Gemini is assigned the identity of “a friendly and helpful AI companion integrated into Waymo’s self-driving cars.” This is not just a simple chatbot integration but a carefully designed intelligent system aimed at “providing useful information and assistance in a safe, reassuring, and unobtrusive manner to enhance passenger experience.”

The system emphasizes concise dialogue—Gemini is instructed to use clear, simple language, avoid technical jargon, and keep responses to one to three sentences. When passengers activate the assistant via the in-vehicle screen, Gemini can select from a set of pre-approved greetings and personalize greetings using the passenger’s name. The system can also access contextual data such as ride history to make interactions more thoughtful.

From Climate Control to Dialogue Boundaries: Gemini’s In-Vehicle Function Design

The revised autonomous taxi assistant system prompt shows that Gemini can access and manage various vehicle functions. According to Wong’s analysis, current supported features include temperature adjustment, lighting control, and music playback—basic needs affecting passenger comfort.

Interestingly, the prompts explicitly list functions that Gemini cannot control—such as volume adjustment, route changes, seat adjustments, and window operations. When passengers request Gemini to perform unsupported functions, the robot responds with a default “expectation-setting phrase,” such as “That’s not something I can do right now,” honestly acknowledging system boundaries while leaving room for future feature expansion.

Waymo spokesperson Julia Ilina told TechCrunch, “While we don’t have details to share today, our team is continuously experimenting with various features to make riding with Waymo enjoyable, seamless, and practical. Some of these features may or may not appear in our passenger experience.” This indicates that Waymo is taking a cautious, iterative approach to feature testing.

Robot Self-Awareness: Differentiating AI Assistant from Self-Driving System

One of the most notable aspects of the system prompts is that Gemini is explicitly instructed to distinguish its identity. The assistant is told to clearly differentiate itself as “Gemini AI Robot” from the autonomous driving system “Waymo Driver.”

This means that when a passenger asks, “How do you see the road?” Gemini should respond with “Waymo Driver uses multiple sensors…” rather than “I use multiple sensors.” This identity distinction reflects Waymo’s clear understanding of responsibility boundaries—the AI assistant’s role is to accompany and serve, not to make driving decisions.

Safety Boundaries and Information Limits: What Gemini Cannot Do

The system prompts include strict restrictions. Gemini is explicitly prohibited from speculating, explaining, confirming, denying, or commenting on real-time driving behavior or specific driving incidents. If a passenger asks about a Waymo accident video or concerns, the robot is instructed to avoid direct answers. The prompts state: “Your role is not to be a spokesperson for driving system performance, and you should not adopt defensive or apologetic tones.”

Additionally, Gemini cannot perform real-time actions such as ordering food, making reservations, or handling emergencies. These limitations reflect Waymo’s cautious stance on safety and responsibility in autonomous vehicle design. In contrast, the assistant can easily handle general knowledge questions—weather queries, Eiffel Tower height, local Trader Joe’s hours, or sports scores.

Technical Benchmarking: Differentiation from Tesla, Cruise, and Grok

This is not Waymo’s first integration of AI into autonomous vehicles. Waymo has already used Gemini’s “world knowledge” to train its self-driving cars to handle complex, rare, and high-risk driving scenarios. However, the way Gemini is applied in autonomous taxis differs significantly from other industry players.

Compared to Tesla’s efforts, Waymo’s design is more pragmatic, focusing on improving the passenger experience with detailed features. Grok, used in some autonomous vehicles, takes a different approach—designed as a long-term onboard companion capable of deep conversations and remembering previous questions’ context. Gemini’s role in Waymo’s autonomous taxis is between these two, providing practical in-vehicle services while maintaining friendly interaction with passengers.

This differentiation reflects different companies’ understanding of the autonomous vehicle ecosystem—Waymo positions its AI assistant as a safe, controllable, boundary-clear service tool rather than an unrestricted intelligent partner. As autonomous vehicle technology matures, the design of onboard AI assistants will become a key factor in user experience differentiation.