Google's Biggest AI Announcements (I Was There)

Google's Gemini 3.5 Flash was described by Logan Kilpatrick as the best model Google has ever shipped and the most widely distributed day-one launch in the company's history, reaching 900 million users across the Gemini app, Google Search, and the developer API. Originally designed as a cheap, smaller workhorse for chat, Flash has been repositioned as the primary model for the agent era, targeting agentic long-running tasks and coding. Kilpatrick placed its intelligence at roughly a Sonnet level, above mini-class models but competitive with much larger models from ecosystem rivals. Gemini 3.5 Pro was announced separately as still in development, with multiple iterations running internally and availability expected the following month.

Gemini Omni was framed by Demis Hassabis as a world model capable of accepting any input type and producing any output type. Google previously maintained separate specialized models for video generation, image generation and editing, text-to-speech audio, and music, and Omni fuses all of these into a single model. The stated goals are a simpler developer experience and cross-pollination of capabilities across modalities. The current Omni release is the Flash variant and is described as only the first iteration. It is launching initially in the Gemini app, YouTube, and Flow, with early API access for developers planned shortly after Google I/O. Kilpatrick predicted Omni will open a new product category around video remixing and editing, and argued it could shift content creation toward more substantive storytelling by removing the bottleneck of limited editor time and storage.

Google launched managed agents in the Gemini API, allowing developers to build agent experiences with a single API call and no orchestration code. A stage demo showed an AI radio show that called approximately seven different models using skills written in Markdown with no orchestration layer. An always-on 24/7 agent inside the Gemini app was also launched, rolling out to trusted testers immediately and to Gemini Ultra customers the following week. MCP support for managed agents was described as coming soon but not yet available at launch, and Kilpatrick characterized the Google I/O release as step one of the managed agent story with roughly 50 additional roadmap items still to be delivered.

Antigravity has evolved from an AI-powered IDE into a full ecosystem that includes an agent manager, IDE, CLI, SDK, and API integration. Google uses Antigravity internally to build its own products, including contributions to the Google Search codebase. It targets agentic engineering for production-quality code and offers flexibility to use other models including the OpenAI API. AI Studio is positioned differently, described as opinionated toward the Google ecosystem and aimed at vibe coding. AI Studio natively supports building Android apps and integrating with Google Workspace, and tens of thousands of people built their first Android app in AI Studio on the first day that feature launched.

Future AI Studio support is planned for Android XR wearables, glasses, watches, and Android Auto app building without writing code, though Kilpatrick acknowledged that native XR and wearable form factor support does not yet work perfectly. He predicted that when glasses launch in the fall, developers will be able to build native XR apps in AI Studio without writing code. Kilpatrick drew an analogy to YouTube, arguing that AI coding tools will enable a generation of solo or small-team software creators in the same way YouTube enabled a generation of content creators.