Google launches Gemini Spark agent

Google unveiled Gemini Spark at I/O on May 19, 2026, adding a cloud-based agent that it says can keep working after a user closes a laptop or turns off a phone. Google describes Spark as a “24/7 personal AI agent” that runs in the background and takes action across a user’s digital life under the user’s direction. The launch extends Google’s push to make Gemini more “agentic” as the company ties the assistant more closely to Gmail, Calendar, Drive and other services. Google’s public product page says Spark is “coming soon,” rather than generally available. ### What, exactly, is Google saying Spark does? Google’s Gemini Spark page says the agent “works in the background 24/7, even if your phone and laptop are turned off.” The page says users can give Spark a task and let it operate autonomously, while choosing when to turn it on. Google also says the system is designed to “check with you before taking major actions,” a safeguard the company is using to frame Spark as automated but not fully unchecked. (blog.google) Google’s broader Gemini app announcement on May 19 said the app was becoming more proactive and capable of handling tasks over time, part of what CEO Sundar Pichai called the “agentic Gemini era” in his I/O keynote. In that keynote transcript, Google linked the new direction to tools that help users do more than generate text, moving toward software that can act on a user’s behalf. (gemini.google) ### Which Google services are tied into it? Google’s own materials place Spark inside the same productivity stack that already underpins Gemini’s consumer and Workspace features. The company says Spark is meant to help users navigate their “digital life,” and reporting on the launch says that includes Gmail, Calendar, Drive, Docs and Slides. Google has already been building toward that model with earlier agent features such as Google Labs’ CC, which connected to Gmail, Calendar and Drive to assemble daily briefings and draft follow-up actions. (blog.google) Android Authority reported after the I/O announcement that Spark could proactively monitor emails, organize notes and create summaries across Google apps. SiliconANGLE similarly described Spark as a 24/7 assistant for daily digital tasks. Those reports align with Google’s own description of a cloud agent that persists beyond a single chat session or device interaction. (gemini.google) ### Why does the privacy question come up so quickly? Google’s design puts Spark close to a user’s personal data. Gmail, Calendar and Drive hold messages, schedules, documents and contacts, and an always-running agent can only be useful if it can read enough of that context to act. Google says users choose to enable Spark and that the system is designed to seek approval before major actions, but the company’s public materials do not spell out all default permissions, retention practices or granular access controls on the main overview page. (siliconangle.com) Forbes reported on May 24 that code findings and product disclosures pointed to usage caps and purchase-related warnings that were not central to Google’s stage presentation. That reporting adds to scrutiny over how much autonomy Spark will have, what counts as a “major” action, and how often users will be asked to approve steps. (gemini.google) ### What changes when an agent is always on instead of chat-based? A persistent agent shifts the engineering burden from a single prompt to an ongoing workflow. Google’s description means Spark may need to watch for triggers, schedule work, retry failed steps and maintain state while a user is offline. That is different from a chatbot that responds once and stops. Google’s I/O materials described Gemini 3.5 as a model family built for “complex, agentic workflows,” language that fits the kind of background execution Spark is supposed to handle. (forbes.com) The cost side also changes. A system that checks inboxes, tracks calendars, drafts outputs and waits for approvals can accumulate compute use over time rather than only when a user opens an app. Google has not publicly detailed Spark pricing on its overview page, but the product’s “coming soon” status suggests those commercial and operational details are still to come. (blog.google) ### How far along is the rollout? Google announced Spark during I/O week, but the company’s product page currently labels it “coming soon.” The Gemini app post and the I/O announcement pages present Spark as part of a broader 2026 push around agents, alongside new Gemini models and developer tools. Google has not, in the materials reviewed, given a firm public launch date for broad availability. (gemini.google) Google’s next concrete step is likely to come through updates to the Gemini product pages or additional I/O follow-up posts. For now, the named participants are Google, the Gemini app team and users who will be asked to opt in when Spark moves beyond its “coming soon” phase. (blog.google)