Sundar Pichai: Gemini 3, Vibe Coding and Google's Full Stack Strategy

Sundar reflects on the successful Gemini 3 and Nano Banana Pro launches, describing how Google has been shipping AI products daily. He explains the decade-long foundation starting from the 2016 AI-first pivot, including the merger of Google Brain and DeepMind, and how the full-stack approach from infrastructure (TPUs) to models to products creates multiplicative effects across the company.

• •Google has been shipping AI products nearly every day recently, demonstrating unprecedented velocity
• •The 2016 AI-first strategy was triggered by AlphaGo (Jan 2016) and first TPU announcement (May 2016)
• •Full-stack approach means innovations at each layer (infrastructure, pre-training, post-training, products) flow through the entire stack
• •Initial capacity constraints required significant upfront investment, making Google appear 'quiet' externally while building foundations
• •Gemini serves as a tangible through-line connecting all Google products from Search to YouTube to Cloud to Waymo

Discussion of Nano Banana Pro's unexpected success with infographics, which wasn't a designed feature but emerged from improved text rendering capabilities. Sundar sees this as unlocking latent creativity in people, giving them tools to express ideas as they envision them rather than being constrained by traditional tools like PowerPoint.

• •Infographics capability emerged organically as text rendering improved, not as a targeted feature
• •The model enables information compression - potentially reversing PowerPoint's trend of expanding slide decks
• •Demonstrates massive latent creativity in the world when people get accessible, expressive tools
• •Grounding with Google Search makes infographics actually useful for organizing and presenting information
• •People can now express ideas the way they think them, removing tool-based constraints on creativity

Sundar shares his personal launch day approach: consuming news and social media to understand user reception, maintaining an unstructured schedule to visit teams, and monitoring dashboards while getting firsthand feedback. He emphasizes the importance of feeling the product reception directly rather than just reading reports.

• •Morning ritual: reads news first, not Google email, since important Google news will appear in media anyway
• •Actively monitors X (Twitter) on launch days to see how average users experience products
• •Keeps schedule flexible on launch days to walk around and engage with shipping teams
• •Combines dashboard monitoring (QPS, usage, capacity) with qualitative feedback from users and teams
• •Values firsthand understanding over reports - needs to 'feel it' directly to assess success

The blue micro-kitchen (MK) in the Grading Canopy office recreates early Google's intimate culture despite the company's massive scale. This space brings together top talent including Sergey, Jeff Dean, Sanjay, and others for spontaneous idea exchange, reminiscent of Google's startup days.

• •The MK makes Google feel small and intimate despite being a massive global company
• •Regularly frequented by Sergey, Jeff Dean, Sanjay, Norm, Coray, Oreo, and other top engineers
• •Dense concentration of talent leads to spontaneous, high-value exchanges of ideas
• •Sundar monitors launch QPS by literally hovering over team members' screens in this space
• •These spaces pull people back to office by demonstrating value of in-person collaboration

Sundar outlines Google's next decade of bets including quantum computing (expecting 'breathless excitement' in 5 years), Project Suncatcher for space-based data centers, robotics work, and continued investments in Waymo, Isomorphic/AlphaFold, and Wing drone delivery. He explains how seemingly crazy ideas like space data centers make sense when considering future compute needs.

• •Quantum computing expected to reach AI-level excitement in approximately 5 years
• •Project Suncatcher aims to build data centers in space with 27 milestones, targeting TPUs in space by 2027
• •Space data centers justified by massive future compute requirements, not current needs
• •Continued long-term bets on Waymo (now reaching inflection point), Isomorphic with AlphaFold, Wing delivery
• •Active development of robotics capabilities as another major future bet

Sundar discusses how AI tools are making coding accessible to non-engineers, similar to how blogs democratized writing and YouTube democratized video creation. He shares examples of Google employees submitting their first code changes and non-technical staff building functional applications, while acknowledging this is 'the worst it will ever be.'

• •Sharp increase in Google employees submitting their first CLs (code changes) due to AI coding tools
• •Example: Communications team member created animated HTML page for teaching Spanish conjugation in one shot
• •Product marketing people now 'vibe coding' prototypes instead of just describing ideas
• •Modern IDEs with AI make coding more enjoyable and approachable, even for experienced developers
• •'This is the worst it will ever be' - capabilities will only improve from current baseline

Sundar expresses excitement for upcoming products including Flow and the growing NotebookLM community. He highlights how NotebookLM has found passionate users among journalists and PhD researchers, demonstrating AI's value for serious research work beyond consumer applications.

• •Flow is a new product Sundar has been personally testing and is excited about
• •NotebookLM has developed a passionate, growing community of power users
• •Journalists and PhD students are using NotebookLM for serious research work
• •Continued Gemini roadmap improvements across all Google products
• •Teams need rest after intense shipping period before next wave of launches