earlier this week, google officially announced that gemini will soon receive a brand-new interface and upgraded automation capabilities. the latest leaked clues suggest that this update may mark a crucial step toward turning gemini into a true “ai agent”—with the launch of a persistent smart assistant codenamed “spark.”
according to multiple sources, a spark toggle has quietly appeared in the menu of the android version of the gemini launcher. according to the official welcome page, spark no longer settles for passive responses; instead, it can proactively execute tasks across apps—automatically clearing invalid emails in gmail, intelligently consolidating calendar entries and notes to generate meeting summaries, or instantly aggregating personalized news briefs based on user preferences.
furthermore, spark supports a “skill-building” feature: users simply define a set of structured commands and inject dynamic variables, enabling them to reuse the same logic to handle similar tasks in bulk—a design philosophy closely aligned with claude projects. it natively supports multi-stage workflow orchestration and can simultaneously retrieve data from multiple applications (currently confirmed to integrate with the google workspace suite). expansion into third-party ecosystems is also under planning.
notably, spark allows users to set up fully automated execution modes, completing end-to-end operations without any manual intervention; in some scenarios, it may even invoke dedicated lightweight ai models for independent computation. additionally, internal screenshots confirm that this assistant has achieved deep control over the chrome browser and can access local device file systems—though it does not yet possess the full-system takeover capabilities of openclaw or claude cowork, its prototype as a “task-oriented intelligent agent” is already clearly visible. as more technical details continue to emerge, google’s strategic roadmap in the ai agent arena is rapidly taking shape.
