Search adds “AI Mode” powered by Gemini 3.5 Flash to answer queries and run background agents

Google on May 19 put Gemini 3.5 Flash at the center of Search’s new AI Mode, extending the company’s latest model from the Gemini app into the interface many users use to reach the web. Google said 3.5 Flash is now the default model for AI Mode in Search globally, alongside its rollout as the default in the Gemini app. (blog.google) The model is designed for what Google calls “agentic” work — tasks that unfold over multiple steps rather than a single answer. Google announced the change at its I/O 2026 developer conference in Mountain View, California. (deepmind.google) ### What exactly changed inside Search? Google said AI Mode is a new Search experience that uses Gemini 3.5 Flash to answer more complex queries and handle multi-step tasks inside Search. (blog.google) In its announcement, the company described 3.5 Flash as the default model for AI Mode and said the model’s speed makes it suitable for long-horizon agentic tasks. Search Engine Land separately reported that Gemini 3.5 Flash began powering AI Mode immediately after the announcement. TechCrunch reported that Google is pitching 3.5 Flash less as a chatbot engine than as a model that can plan, execute and iterate across tasks. Ahead of the launch, DeepMind Chief Technologist Koray Kavukcuoglu told reporters the model combines “quality and low latency” and is built for coding, multimodal reasoning and agentic workloads. (blog.google) ### How much context does the system get? Google DeepMind’s model card says Gemini 3.5 Flash accepts text, images, audio and video, with a context window of up to 1 million tokens and a 64,000-token output limit. Google lists Google Search AI Mode among the distribution channels for the model, alongside the Gemini app, Gemini API and enterprise products. (blog.google) Google’s product pages for Gemini 3.5 Flash describe the model as suited to “long horizon tasks” and multi-step problem solving. The company also says the model is four times faster than other frontier models when measured by output tokens per second, though that performance claim comes from Google’s own announcement materials. (techcrunch.com) ### What are these “background agents” supposed to do? Google’s I/O materials and follow-on coverage describe AI Mode as moving beyond one-off answers toward persistent task handling. Third-party reports summarizing the I/O demos said Search agents can monitor topics in the background and return updates or alerts, while Google’s own materials frame 3.5 Flash as a model for extended agent workflows. (deepmind.google) Inc. reported that Search chief Liz Reid introduced AI-driven changes to Search at I/O, including a more capable search box and tools aimed at handling more detailed requests. Yahoo Finance, citing Google’s presentation, described examples in which users searched across photos, videos, files and Chrome tabs rather than only the open web. (blog.google) ### Why does this matter for websites and publishers? Google’s move places its newest AI model in front of the retrieval step itself, not only in a separate assistant product. Time reported that Google said the search bar would be “completely reimagined with AI,” while USA Today described the update as Google’s biggest Search change in more than 25 years. (androidheadlines.com) That means more user queries may be answered, refined or routed inside Google’s AI layer before a user clicks through to a site. Search industry research has already documented broad expansion of AI-generated answers in Google results. One March 2026 data point cited by multiple search-marketing reports put AI Overviews on about 48% of Google queries, up from 34.5% in December 2025, though those figures come from third-party measurement firms rather than Google. (inc.com) ### What comes next from Google? Google said at I/O 2026 that Gemini 3.5 Flash is available now through AI Mode in Search, the Gemini app, Google AI Studio and enterprise products. The company also said Gemini 3.5 Pro is already in internal use and is expected to roll out next month, giving Google another model release to watch after the Search changes announced on May 19. (time.com) (blog.google) (marketing-tutor.com)