Google I/O agentic AI gets serious
- Google used its May 19 I/O 2026 conference to push agentic AI beyond consumer demos and into scientific research workflows, with new Gemini for Science tools. - Google DeepMind said Hypothesis Generation uses a multi-agent system simulating the scientific method, while EDN called this year’s I/O messaging “maturity.” - Researchers can access experimental tools through Google Labs and Science Skills in Google Antigravity, according to Google’s May 2026 product pages.
Google used its I/O 2026 conference on May 19 to present agentic AI as a working layer for research, not only as a consumer software feature. The company introduced Gemini for Science, a set of experimental tools aimed at helping researchers synthesize literature, generate hypotheses and run computational discovery workflows, according to Google product pages. MIT Technology Review reported on May 22 that the announcements showed Google trying to build systems that assist scientific research more directly, extending beyond earlier showcase moments tied to headline breakthroughs. EDN, in a separate May 21 article, said this year’s I/O presentations showed more “maturity” than prior rounds of AI messaging. ### Which Google announcements made this feel different from a standard I/O AI pitch? Google’s I/O roundup said on May 19 that the company had “moved beyond AI tools that just help us write, to agents that help us act,” linking that shift to updates in Google Antigravity, its agent-focused development platform. Sundar Pichai used the same event to describe I/O 2026 as part of an “agentic Gemini era,” framing the company’s latest product cycle around systems that can execute multi-step tasks. (blog.google) Google DeepMind’s science push sat inside that broader product agenda. The Gemini for Science page describes a package of “experimental tools” and “Science Skills” meant to support scientific work rather than general-purpose chat or search. ### What, exactly, is Gemini for Science supposed to do? Google’s product page lists three main tools in Google Labs: Literature Insights, Hypothesis Generation and Computational Discovery. Literature Insights is described as a way to synthesize scholarly literature and extract paper data into tables mapped to source evidence. (blog.google) Hypothesis Generation is described as a multi-agent system that simulates the scientific method to identify knowledge gaps and propose testable research plans. Computational Discovery is described as an agentic research engine that generates and scores code variations against user-defined optimization metrics. (ai.google) Google also said researchers can use Science Skills in Google Antigravity to orchestrate models and scientific resources in a “professional scientific workbench.” The company said those workflows are meant to compress multi-step analysis pipelines that would otherwise take hours. ### How does this connect to Google DeepMind’s research work? Google DeepMind said on May 19 that it published Co-Scientist research in *Nature* and made that system available to individual researchers through Hypothesis Generation. (ai.google) The company described Co-Scientist as a multi-agent AI partner built with Gemini that “iteratively generates, debates, and evolves” hypotheses for complex scientific problems. Google’s science pages also tied the tools to named research examples. The company said Ben Luisi’s lab at the University of Cambridge used its AI tools on antimicrobial-resistance work, Duke University’s Wang Lab used Deep Think mode on crystal-growth fabrication methods, and Rutgers mathematician Lisa Carbone used Deep Think mode to review a specialized mathematics paper. ### Why did outside coverage focus on “maturity”? MIT Technology Review wrote on May 22 that Google’s path in AI-driven science is shifting from highly specialized breakthrough systems toward tools that can more directly assist researchers. (deepmind.google) That characterization matched the product mix Google showed at I/O: less emphasis on a single moonshot result, more emphasis on software that can be used inside everyday research workflows. (ai.google) EDN’s Brian Dipert wrote on May 21 that what distinguished this year’s event from earlier messaging was “maturity.” His article placed Google’s I/O announcements inside a broader move from agentic AI as an idea to agentic AI as deployed product behavior. ### Where does Google go from here? Google’s May 2026 science materials say the next step is researcher use through Google Labs and Google Antigravity, where the company is exposing Literature Insights, Hypothesis Generation and Computational Discovery as experimental products. (technologyreview.com) Google DeepMind’s May 19 Co-Scientist post said the system is being made available to individual researchers through Hypothesis Generation, providing the clearest near-term milestone for whether the I/O science push moves beyond conference demos. (edn.com) (ai.google)