Our Field Trip to Google I/O + A Sit-Down With Sundar Pichai + System Update

What It Covers

Kevin Roose and Casey Newton attend Google I/O 2026, reporting on Gemini 3.5 Flash, agentic search, and AI coding tools, then sit down with Google CEO Sundar Pichai to discuss Google's competitive position, AGI timelines, AI public perception, and government regulation of frontier models.

Key Questions Answered

• •Google's competitive gap: Sundar Pichai openly acknowledges Google trails rivals specifically in agentic coding, long-horizon tasks, and instruction following — not across all AI capabilities. Practitioners evaluating AI coding tools should weight Claude Code and Cursor-adjacent products over Gemini for complex, multi-step codebases until Google ships Gemini 3.5 Pro, expected within roughly 30 days.
• •Speed-and-cost strategy over frontier quality: Google is betting on Gemini 3.5 Flash being four times faster and cheaper than competing frontier models rather than achieving top benchmark scores. For companies burning billions of tokens daily, this efficiency advantage matters. Individual power users seeking best-in-class reasoning output should still default to higher-tier models from Anthropic or OpenAI.
• •AI agent trust-building follows autonomous vehicle logic: Pichai frames agent adoption as a stepwise trust process, mirroring how self-driving cars required gradual passenger confidence-building. Practically, this means deploying agents on low-stakes, reversible tasks first — calendar color-coding, meeting prep — before granting access to consequential workflows, financial data, or external communications.
• •AI public perception gap is structural, not messaging-driven: A New York Times and CNN poll found only 16% of Americans view AI as mostly good, while 35% view it as mostly bad. Pichai attributes this to humans being poorly equipped to process rapid technological change, suggesting companies should prioritize demonstrating concrete productivity gains over broad awareness campaigns to shift sentiment.
• •Recursive self-improvement requires external governance, not internal protocols: Pichai states that reaching true recursive self-improvement would demand industry-wide coordination beyond any single lab's internal decision-making. Organizations tracking AGI risk should monitor whether labs establish cross-industry governance frameworks before this threshold, as Pichai explicitly warns against competitive race conditions at advanced capability levels.