Google pushes ambient AI at I/O

Google used I/O on May 19 to present a broader shift than a model launch. The company introduced Gemini 3.5, unveiled Gemini Omni and expanded Antigravity, its agent-development platform, while executives described AI as something that can act across products and persist through workflows rather than sit inside a single app feature. Google CEO Sundar Pichai said the company was entering the “agentic Gemini era,” and Google Cloud tied that message to enterprise products including Gemini Enterprise, Workspace and Agent Platform. That framing matters because Google was not only selling a faster model. Google’s own launch materials described Gemini 3.5 as a family that combines “frontier intelligence with action,” and described Gemini Spark as a “24/7 personal AI agent” for enterprise users. In other words, the company’s pitch at I/O was that AI should become an always-available layer spanning devices, services and work tasks. (blog.google) ### What did Google actually launch at I/O? Google said at I/O 2026 that it was releasing two new models, Gemini Omni and Gemini 3.5. In its official conference roundup, the company said Gemini Omni can “create anything from any input, starting with video,” while Gemini 3.5 is the first model in a new family built for action-oriented use. (blog.google) Google DeepMind said Gemini Omni Flash is the first model in the Omni family. The company said it is rolling out to the Gemini app, Google Flow and YouTube Shorts, with image and audio outputs planned later. Google’s model card says Omni Flash can generate and edit high-resolution video from varied inputs and can simulate real-world physics. (blog.google) ### Why does “ambient” matter more than a single model name? Sundar Pichai said in his keynote transcript that Google was moving into the “agentic Gemini era.” Google Cloud used similar language, saying Antigravity’s expanded capabilities and integration with Agent Platform would bring agentic development across an organization. (blog.google) Google’s product pages show how that idea is being packaged. Gemini 3.5 is described as handling “complex, agentic workflows,” while Spark is positioned as an always-running assistant that can take actions under a user’s direction. That suggests Google is trying to define AI less as a chatbot session and more as software infrastructure that remains available across contexts. That last point is an inference from Google’s product descriptions and launch language. (blog.google) ### Where do world models fit into this push? Google DeepMind said Gemini Omni is a “leap in world understanding, multimodality, and editing.” Its product page says the model combines an understanding of physics with Gemini’s broader knowledge and can use image, text, video and audio as references for a single output. The model card for Gemini Omni Flash adds that the system was developed with internal safety, security and responsibility teams and underwent evaluations and red teaming. (blog.google) That language places Omni not only in the creative-model category but also in the class of systems Google says are meant to reason about scenes, motion and physical consistency. (deepmind.google) ### Which Google products are carrying this message into deployment? Google Cloud said the I/O announcements were being pushed directly into Gemini Enterprise and Google Workspace. The company paired Gemini 3.5, Gemini Omni, Antigravity and Spark inside the same enterprise launch post, linking the consumer keynote to workplace software and developer tools. (deepmind.google) Google’s main I/O collection page also grouped the announcements around models, agents and tools rather than around one flagship app. That packaging supports Google’s effort to show AI as a cross-product layer, with model releases, agent orchestration and workplace automation presented as parts of one stack. That is an inference from how Google organized the announcements. (cloud.google.com) ### What questions does this leave for rollout? Google DeepMind’s model materials emphasize safety reviews, red teaming and edited-content transparency efforts. The company also published model cards for Gemini Omni Flash and Gemini 3.5 Flash on May 19, giving developers and customers a formal record of capabilities and limitations. (blog.google) The next concrete step is broader product rollout. Google said Gemini Omni Flash is already reaching the Gemini app, Google Flow and YouTube Shorts, while Gemini 3.5 and Antigravity are being positioned for developers, Workspace users and Google Cloud customers following the May 19 I/O announcements. (blog.google) (deepmind.google)