Most AI assistants wait for you to start a conversation. Google's new Gemini Spark, announced at I/O on May 19, is built around the opposite idea.
Spark is a personal AI assistant built on Gemini's base models with Gmail integration and 24/7 operation - it runs in the background rather than only responding when you open the app and type something. Google built it using an "agentic harness" - software architecture that gives the AI structured, permission-gated access to external tools and services - developed by an internal team called Google Antigravity.
Gmail is the launch integration: Spark monitors your inbox, surfaces what matters, and can take actions like drafting replies or flagging follow-ups without requiring you to prompt it each time.
The Trade-off Users Will Care About
A persistent AI with ongoing inbox access is a real trust decision. The utility case is clear - fewer missed follow-ups, faster triage on a busy inbox, less time spent on email management. The concern is equally clear: Google already stores email for hundreds of millions of users, and giving an AI agent the authority to act on that data is a different category of permission than passive storage.
Spark enters a market where similar background-running assistants are being developed at Anthropic, OpenAI, and a range of startups. Google's practical advantage is integration depth: connecting Spark to Gmail, Calendar, and Drive requires no setup from users who already rely on those services. Third-party assistants attempting the same thing require OAuth flows, separate permission grants, and ongoing maintenance.
No pricing or general availability timeline was announced at I/O.