Inside Google's AI turnaround: AI Mode, AI Overviews, and vision for AI-powered search | Robby Stein
Robby Stein, VP of Product at Google, shares insights on the development of Google's AI products, including AI Mode and AI Overviews. He discusses the product principles that guided the creation of billion-user products like Instagram Stories, the philosophy of "relentless improvement," and why AI is expanding, not replacing, Google Search.
The New Era of Google AI and Search
Google’s recent momentum in AI, exemplified by the Gemini app reaching number one in the App Store, is not due to a single change but rather an “incredible sense of focus and urgency” and the compounding effect of continuous, ruthless improvement. The technology has hit a tipping point where the models can now truly deliver value for consumers.
Contrary to the narrative that AI chatbots would make traditional search obsolete, AI is proving to be expansionary. The core needs for Google Search—finding specific, factual information like a phone number or directions—remain unchanged. However, AI is unlocking the ability to answer a new, wider set of complex questions and fulfill user curiosity in novel ways. This is evident in the growth of multimodal search, with Google Lens seeing a 70% year-over-year increase in visual searches.
Inside AI Mode and AI Overviews
AI Mode is designed as a frontier search experience, created specifically for complex informational needs. Its unique power comes from its ability to tap into Google’s vast, structured, and real-time data, such as the 50 billion products in the Google Shopping Graph and 250 million places in Maps. This is fundamentally different from other chatbots.
Under the hood, when a query is made in AI Mode, the model performs a “query fanout,” using Google Search as a tool to run dozens of background searches. It then synthesizes the information, leveraging core Google signals for quality, authoritativeness, and spam detection to construct its response. This is why the best practices for creating helpful, original, and well-sourced content still apply for creators aiming to appear in AI-generated answers.
The development of AI Mode took roughly one year from concept to a major public launch. The journey began with the observation that users were trying to ask more complex, natural language questions in AI Overviews. A small, entrepreneurial team of 5-10 people was formed to explore a dedicated experience. The conviction to invest further came not from a metric, but from a qualitative “moment of brilliance” when an early prototype demonstrated its power. This led to a trusted tester program, a broader release in labs for scaled feedback, and finally a full launch.
Core Product Principles
Stein outlines three core principles for building successful products, plus a critical addendum:
-
Deeply Understand People: Go beyond surface-level user feedback and understand the “job” a user is “hiring” your product to do, a concept from the “Jobs to Be Done” framework. Be a student of causation to understand the underlying motivation that leads a person to your product. This focus on the “why” allows you to think from first principles rather than being constrained by your product’s current tools.
-
Maintain Analytical Rigor: Use data to dissect problems and get to the true root cause. When a product or feature is failing, you must be able to understand precisely where and why. This was crucial in diagnosing the initial failure of Instagram’s “Close Friends” feature, where data showed it was being mistranslated as “Best Friend,” leading users to create lists that were too small for the feature to be effective.
-
Design for Clarity Over Cleverness: Leverage established patterns and mental models that users already understand. Reinventing standard conventions, like using a non-standard icon for a camera, adds cognitive load and friction. As Don Norman detailed in The Design of Everyday Things, a product’s design should clearly communicate its function.
-
Coda: Be Humble: Constantly question yourself, listen to users and colleagues, and always be open to being wrong.
Lessons from Instagram: Stories and Close Friends
The decision to build Instagram Stories was driven by understanding the core “job” of Instagram—“sharing your life and connecting with people.” Snapchat had pioneered a format that was ephemeral, full-screen, and lower-pressure, which was a fundamentally better way to fulfill that job for certain moments.
“Not every great thing is going to be invented by you… At the end of the day, you’re kind of just robbing your user base of the opportunity to have a better product if you’re not making the best possible product for your use cases.”
The key was to adapt the format to make it feel native to Instagram. This included adding unique creative tools, allowing camera roll uploads (a key friction point in other apps), and adding the ability to pause a story.
The “Close Friends” feature was a multi-year project that failed completely at first. The initial implementation was confusing, spanning both Feed and Stories. The breakthrough came from applying the core product principles:
- Understanding the Job: The goal wasn’t just to share privately; it was to feel an emotional sense of connection. The product only succeeded if the user got a DM back.
- Analytical Rigor: Data showed that the feature worked for users who added 20-30 people to their list, as this critical mass ensured a reply.
- Designing for Clarity: The UI was changed to put a green ring around the outside of the story bubble, making it immediately clear to the viewer that this was special content for them.
On Relentless Improvement and Team Size
Stein’s personal mantra is to “embody relentless improvement,” which combines constant, productive effort with a deep-seated dissatisfaction with the status quo. It’s about refusing to habituate to problems and constantly asking how things can be better.
He also cautions against the “cult of lean,” where teams are kept small for too long. While it’s right to start small to find initial conviction, many potentially transformative products die because they are under-resourced. Once there is conviction, you must invest enough to build a truly great version of the product. The long development time for “Close Friends,” for instance, was partly because the team was kept too small for too long.