Technology

Google Turns Its Spark Agent Into Free Tasks Across Six Markets

Martin HollowayPublished 24m ago3 min readBased on 2 sources
Reading level
Google Turns Its Spark Agent Into Free Tasks Across Six Markets
source:blog.google

Google is renaming its Spark AI agent to Tasks and making it free for all users in the US, Canada, Japan, Korea, Australia, and India. The Verge report, dated November 17, 2026, describes the change as a rename of the agent Google first introduced in May. Until now, Spark was limited to subscribers of Google AI Pro and Ultra.

Tasks runs on Google's Gemini 4 Hydrogen model. Users reach it inside the Gemini app by switching the homepage toggle from Chat to Tasks. Google positions the agent as a task executor rather than a conversational assistant, and the feature set reflects that.

In every market covered, Tasks can build meal plans, manage the user's calendar, and search email to identify recurring expenses. In the US, it can also call businesses to compare product pricing and check inventory, buy items with Google Pay, and sign into accounts that support Google Password Manager. US users can assign work to Gemini by emailing gemini@google.com or sending a text message, so the agent can be dispatched from outside the app.

Google's description of the execution environment is the detail most relevant to security teams. According to Google, Gemini performs tasks in a hardware-isolated virtual machine (VM) dedicated to the user's account, and that machine is never shared with other users. For sensitive actions, such as purchasing an item, Gemini stops and asks for permission before completing the step.

Google's AI blog carries a post titled “Tasks in Gemini are now free of charge, with new ways to get things done on your behalf.” The post credits the announcement to Josh Woodward, Vice President, Labs, Gemini app, and AI Studio.

The broader context here is the architecture of trust. An agent that signs into accounts through a password manager, places outbound calls to merchants, and spends money through a payment rail works with credentials and permissions designed for a person at a keyboard. Google's answers are per-account isolation at the VM level and a confirmation gate on purchases. Both are claims Google has made, and neither has been independently assessed. The questions an expert would ask next include how the permission boundary behaves on ambiguous requests, what audit trail the user receives, and how a compromised email or text channel would be handled as an input path for commands.

In my view, removing the subscription requirement matters as much as the rename. Agentic features tied to a paid tier reach a small, self-selected group of users. Moving the same capabilities to a free tier in six markets puts the confirmation flow and isolation model in front of a much larger population. That scale will test whether the permission prompts are specific enough to be useful, or whether people learn to approve them without reading.

The practical upside is concrete. Meal planning, calendar triage, expense reconstruction from years of inbox history, and price and stock checks at local shops are chores that consume attention without requiring judgment at each step. If the permission gate holds and the isolation claims survive scrutiny, an agent that handles this work reliably from a phone or a text message becomes a useful layer on top of existing accounts, rather than another app users have to learn.