Google is preparing to roll out its most ambitious AI agent yet, called “Gemini Spark.”
The always-on assistant aims to challenge Anthropic’s Claude Cowork by automating multi-step tasks in apps without any manual oversight.
This new feature, found by tinkering with the latest beta (version 17.23) of Google’s app, allows users to have an “Agent” tab in Gemini. Rather than acting like regular chatbots that only answer questions, Spark is able to perform tasks like cleaning Gmail spam, compiling meeting summaries from various documents, and generating personalized news summaries itself.
Leaks suggest that Spark lives in Gemini’s overflow menu. Once activated, it can create “skills” to automate recurring tasks, just like Claude’s Project functionality.
Spark works with information collected from apps, conversations, activities, geographic information, and even related websites when the user is logged in.
Advanced capabilities may include controlling the Chrome browser as a self-driving car and being able to access documents on devices. But early signs show that Spark cannot yet control an entire computer, unlike OpenClaw and Claude Cowork.
Despite substantial results, Google warns that the model remains “experimental”. The disclaimer can be seen in the beta notes: Even though Spark asks for permission before sensitive actions, it may have chances to share information or make purchases without user permission.
