What is Gemini Spark? Google's new AI agent leaks days before I/O

Google is preparing to roll out its most ambitious AI agent yet, referred to as “Gemini Spark.”

The always-active assistant is aimed at challenging Anthropic’s Claude Cowork by automating multi-step tasks across apps without any manual oversight.

This new feature, which was found by tinkering around with the newest beta version (version 17.23) of Google’s app, allows users to have an “Agent” tab inside Gemini. Rather than acting like ordinary chatbots that only answer questions, Spark is able to do tasks such as cleaning up Gmail spam, compiling meeting summaries from various documents, and generating customised news summaries on its own.

Leaks suggest that Spark lives inside Gemini’s overflow menu. Once turned on, it can create “skills” for automating recurring tasks, just like Claude’s Project feature.

Spark works with information that is collected from related apps, conversations, activities, geographic information, and even websites when the user is logged in.

Advanced abilities might include controlling the Chrome browser as a self-driving car and being able to access documents on devices. But the early signs show that Spark cannot control a whole computer yet, unlike OpenClaw and Claude Cowork.

Despite offering substantial results, Google warns that the model remains "experimental.” The disclaimer is visible in the beta notes that while Spark requests permission before sensitive actions, it may have chances of sharing information or making purchases without the user’s permission.