Google pushes agentic AI and counters spam

Google used its I/O conference on May 19 and May 20 to present what it called a broader “agentic” AI push across Search, Gemini and developer tools. The company said the updates are meant to move users from one-off prompts toward systems that can carry out multi-step tasks. Outside coverage after the event focused on two immediate questions: whether consumers understand what Google is selling, and whether AI-generated answers can be manipulated by businesses trying to steer results. Google’s own announcements paired the new agent language with more changes to Search, where the company still makes much of its money. ### What exactly did Google unveil at I/O? Google said on May 19 that AI Mode in Search would now use Gemini 3.5 Flash as its default model globally. The company also described a redesigned Search box, more agent features inside Search, and broader efforts to make Gemini act on behalf of users rather than only answer questions. In a separate I/O roundup, Google said it had introduced new models including Gemini Omni and Gemini 3.5, and framed the event around what it called an “agentic Gemini era.” (blog.google) Sundar Pichai, Google’s chief executive, used his keynote to describe a year of rapid product releases and to position Gemini as a tool for getting more done across Google’s services. Google’s product posts also highlighted a more proactive Gemini app, new developer tooling under the Antigravity name, and additional agent features for builders. (blog.google) ### Why is Google centering this on Search? Axios reported on May 21 that Google is trying to adopt AI aggressively without damaging the search and advertising businesses that generate most of its profit. That tension has shaped the company’s product rollout: Google is adding conversational and agent features inside Search rather than abandoning Search as the main consumer entry point. (blog.google) Google’s own Search post made the same case in product terms. The company said users can move from traditional queries to more conversational interactions, and from there to agent-driven tasks, all inside Search. That keeps AI usage tied to Google’s existing distribution advantage while competitors such as OpenAI and Anthropic push standalone assistants and chat products. The comparison to rivals is drawn from Axios’ reporting, not from Google’s announcement. (axios.com) ### Why are some early reactions calling the pitch confusing? TechCrunch reported on May 21 that Google’s consumer-facing agent story was one of the most promising parts of I/O but also one of the hardest to understand. The publication said Google was pitching a new way to use the web through AI agents, while leaving open how ordinary users would distinguish among the company’s growing set of AI products and modes. (blog.google) EDN, also writing after I/O, said this year’s conference showed “agentic AI” moving from a slogan to a more concrete product framing. That assessment suggested Google had become more specific about how agents fit into Search, apps and development tools, even as the broader consumer case remained unsettled in early coverage. ### What is the spam problem around AI answers? (techcrunch.com) BBC reporting cited in the source briefings said Google has been working against attempts by businesses and spammers to manipulate AI-generated results. The same briefing said similar anti-spam signals have appeared in responses from ChatGPT and Claude, pointing to a wider problem for AI systems that summarize or rank information from the web. I could not independently retrieve the BBC article text through search results here, so that characterization relies on the cited briefing and secondary references in the prompt. (edn.com) Google’s product strategy makes that issue more sensitive because AI Mode and related features increasingly sit closer to the main Search experience. If users rely on generated answers for recommendations, shopping or factual summaries, any successful effort to game those outputs becomes a product-quality problem for Google, not only a spam problem for the open web. That is an inference from Google’s Search rollout and the reported manipulation concerns. (axios.com) ### What should readers watch next? Google’s next public checkpoints are already listed on its I/O and product pages. The company said Search updates began rolling out starting May 19, while separate posts for Gemini, model releases and developer tools describe additional launches tied to the conference and subsequent product updates. (blog.google) Google I/O 2026 was held May 19-20 at Shoreline Amphitheatre in Mountain View, California, and Google’s post-event pages now serve as the company’s running record of what ships next. For this story, the clearest places to track follow-through are Google’s Search updates page, its Gemini announcements, and outside coverage testing whether consumers adopt the new agent features. (blog.google 1) (blog.google 2)