Google used its annual I/O developer conference on Tuesday to unveil the most sweeping transformation of its product lineup in years, placing agentic artificial intelligence at the centre of everything from search and shopping to coding, smart glasses, and the physical world.
The centrepiece of the event was Gemini 3.5 Flash, described by DeepMind chief technologist Koray Kavukcuoglu as the company’s strongest model yet for coding and autonomous tasks. Running at twelve times the speed of comparable frontier models, Flash can independently manage research projects, execute multi-step coding pipelines, and — in internal tests — build an operating system from scratch inside Antigravity, Google’s agentic development platform. The model is now the default across the Gemini app and Search globally.
Built on that foundation, Google introduced Gemini Spark, a 24/7 personal AI agent that runs on cloud-based virtual machines and requires no open laptop to keep working. Described by CEO Sundar Pichai as a tool that navigates users’ digital lives under their direction, Spark integrates natively with Gmail, Docs, Sheets, and Slides, and can be emailed directly or monitored on Android via a new interface called Halo. It will be available to Google AI Ultra subscribers next week.
The multimodal ambitions of the company were embodied in Gemini Omni, a new model family that reasons across text, images, audio, and video to generate consistent, knowledge-grounded outputs. A demonstration showed it rendering a claymation protein-folding explainer from a single prompt. Omni Flash begins rolling out today to YouTube Shorts and the Flow creative studio, with a more powerful Pro version to follow.
Google also connected its Street View archive — 280 billion images across 110 countries collected over twenty years — to Project Genie, its interactive world model. The integration allows Genie to simulate real locations with adjustable conditions such as weather or time of year, a capability already being used by Waymo to train its autonomous vehicles on rare events. DeepMind research scientist Jack Parker-Holder said physics awareness, currently absent, was roughly six to twelve months behind the quality of video generation.
Across its productivity suite, Google pushed voice as the primary new input layer. Workspace apps including Docs, Keep, and Gmail now accept spoken multi-step instructions, with Gemini handling mid-sentence changes of mind. Google Keep will structure rambling voice notes into organised lists automatically, while Gmail users can ask conversational questions about their inbox.
On the commerce side, Google launched Universal Cart, a centralised AI-powered shopping hub spanning Search, YouTube, and Gmail that tracks price drops, flags product compatibility issues, and surfaces hidden savings. Alongside it, the company detailed its Agent Payments Protocol, which will allow AI agents to complete purchases autonomously within user-defined spending limits — a move that positions Google as a potential intermediary across the entire consumer buying journey.
For developers, Antigravity 2.0 arrived as a standalone desktop application enabling parallel agent orchestration, scheduled background tasks, and a new command-line interface. Google AI Studio now supports native Android app creation in minutes rather than weeks, opening development to non-technical creators. An Android CLI tool, stable at version 1.0, extends these capabilities to third-party agents including Claude Code and OpenAI’s Codex.
Rounding out the announcements, Google revealed partnerships with Warby Parker and Gentle Monster to produce AI-powered audio glasses, designed with Samsung and compatible with both Android and iOS, available later this year.
The scale of Tuesday’s releases made clear that Google is no longer positioning Gemini as a chatbot competing with ChatGPT. It is repositioning the entire company around AI systems that plan, act, and persist — with or without a human watching.
