Google Launches Gemini Spark AI Agent and Gemini 3.5 Models

Google has introduced Gemini Spark, a new "agentic" AI assistant capable of performing proactive tasks on behalf of users by operating in the background across its product ecosystem [1]. The tool, which functions as a 24/7 digital assistant, was announced alongside the rollout of the Gemini 3.5 Flash model during the company’s annual Google I/O developers conference [1, 3].

Key takeaways

• Gemini Spark is a cloud-based AI agent designed to handle routine tasks like sorting emails and meeting notes even when a user’s device is locked [1, 3].
• Gemini 3.5 Flash, the new default model for the Gemini app and Google Search, is described by the company as its strongest agentic and coding model to date [1, 2].
• Google is introducing a "Universal Cart" shopping tool that tracks price history and deals across various services, including YouTube and Gmail [1].
• To address safety concerns, Gemini Spark is programmed to request user permission before executing "high-stakes" actions such as making purchases or sending emails [1].
• Google reported that monthly active users for the Gemini app have surpassed 900 million, more than doubling over the past year [1].

Expanding the Agentic Era

The launch of Gemini Spark marks a shift toward "agentic" AI, a category of technology that moves beyond traditional chatbots by taking action under user direction [1, 2]. Powered by the Gemini 3.5 Flash model and the Antigravity platform, Spark is designed to integrate deeply with tools like Gmail, Docs, and Slides to manage complex workflows [3]. While the agent is capable of working autonomously in the cloud, Google has implemented safety protocols requiring explicit authorization for significant tasks [1]. Select testers will gain access to the agent starting this week, with a broader beta rollout planned for U.S.-based Google AI Ultra subscribers [1, 2].

Alongside the agent, Google is updating its core infrastructure with the Gemini 3.5 family. The Flash model, which is now the default for the Gemini app and "AI mode" in search, is noted for its speed and improved safety training, which the company claims reduces the likelihood of generating harmful content or refusing safe queries [1, 2]. Furthermore, Google announced Gemini Omni, a model capable of creating and editing realistic video by understanding physical forces like gravity and fluid dynamics [1, 2].

Why it matters

Google’s aggressive push into agentic AI comes as the company faces increased competition and pressure from investors to demonstrate the value of its massive capital expenditures, which may reach $190 billion this year [1, 2]. By moving from simple conversational chatbots to agents that can navigate digital lives and manage shopping through the new "Universal Cart," Google aims to deepen its integration across the daily routines of its 900 million Gemini users [1, 2]. As the company navigates this transition, it is also expanding its content verification tools, such as SynthID, to help users distinguish between AI-generated and human-captured media [1].