Google unveils Gemini Spark agent

Google on May 19 used its I/O 2026 keynote to introduce Gemini Spark, a new personal AI agent built into the Gemini app. The company said the product is designed to run continuously, help users manage digital tasks and take actions across apps and the web under user direction. Google tied the launch to Gemini 3.5 Flash, its new model tuned for faster “agentic” work, and said Spark would begin with a limited rollout to trusted testers before a broader beta in the United States. ### What exactly did Google announce at I/O? Sundar Pichai said in his May 19 keynote that Google was entering an “agentic Gemini era,” framing the company’s latest product push around software that does more than answer prompts. In a separate Google post published the same day, the company said Gemini Spark is “your personal AI agent” and can take action on a user’s behalf while remaining under that user’s direction. (blog.google) Google’s product post said Spark is part of the next stage of the Gemini app, which the company described as becoming more proactive and persistent. The same post said more than 900 million people in 230 countries and territories use Gemini each month, up from 400 million users cited by Google at last year’s I/O. (blog.google) ### How is Gemini Spark supposed to work day to day? Google said Gemini Spark runs “24/7,” helping users “navigate your digital life” and complete tasks across services. The company’s description presents the agent as software that can plan and act over longer stretches rather than respond one prompt at a time. (blog.google) Varun Mohan and Logan Kilpatrick, writing in a Google I/O developer post, said Google is “accelerating the shift from prompts to action” with Gemini 3.5 Flash. They said the model is intended for “real-world agentic workflows,” linking Spark’s consumer product pitch to Google’s broader effort to build tools for longer, multi-step tasks. (blog.google) ### Why does this matter for apps and online services? Google’s own language points to a change in how users may reach digital services. If an agent can invoke tools, move across apps and complete tasks on behalf of a user, then the app interface is no longer the only place where discovery and conversion happen. That is an inference from Google’s product description and developer framing, not a separate Google statement. (blog.google) The implication for developers is that compatibility with agent workflows may matter alongside traditional app-store tactics. Google’s I/O materials also highlighted “agent-first” development tools, including Google Antigravity and managed agents in the Gemini API, showing the company is pairing the consumer launch with software infrastructure for developers. (blog.google) ### What model is underneath Spark? Google said Gemini Spark uses Gemini 3.5 Flash, which it described as the first model in its next generation of systems combining high-end performance with lower latency. In its May 19 post, Google said 3.5 Flash is available through the Gemini app, AI Mode in Search and Google’s developer tools. (blog.google) Google also said 3.5 Flash is suited to “long-horizon agentic tasks.” In the same announcement, the company said Gemini 3.5 Pro is being used internally and is expected to roll out next month. ### Who gets access first, and when? Google said on May 19 that Gemini Spark was starting to roll out to trusted testers immediately. (blog.google) The company also said it plans to bring a beta to Google AI Ultra subscribers in the United States next week. Google I/O 2026 runs May 19-20 online, according to the event page, with additional sessions and on-demand materials scheduled to follow. (blog.google) That gives developers and subscribers the next concrete checkpoints: session details on the I/O site and the planned U.S. beta rollout for AI Ultra users next week. (io.google)