News 3 min read

Google makes Gemini Tasks free: its agent can call, shop and book


Google is making Tasks, the agent mode in the Gemini app, free for “hundreds of millions of people” in the U.S., Japan, Korea, Canada, Australia and India, Josh Woodward, Google’s VP for Labs, the Gemini app and AI Studio, wrote on November 17, 2026. Tasks runs on Gemini 4 Hydrogen and can call businesses, check out with Google Pay and sign in to websites for you. Until now it was limited to testers and U.S. Google AI Ultra subscribers: TechCrunch notes it launched in May as Gemini Spark for select testers before reaching Ultra.

What’s new

In the announcement, Tasks is a switch on the Gemini home screen: flip it from Chat to Task and hand Gemini a job with several steps instead of a question. Google lists these things it can do:

  • Phone calls. Gemini calls businesses and books appointments, such as a car service, and adds them to Google Calendar. Calling is not available in Indiana, Louisiana, Minnesota, Montana and Nebraska.
  • Bills and email. It can read utility bills and emails to find what is driving a high bill, using Google Password Manager to sign in to the utility’s portal.
  • Shopping. It compares products with Google’s Shopping Graph, looks for discount codes and checks out with Google Pay. Payments go through the Agent Payments Protocol (AP2) with Visa and Mastercard digital tokens, with American Express “coming soon”. You can set a spending cap, and receipts land in Google Wallet, which is now expanding to iOS.
  • Groceries. Adds items to your Instacart order, including from a photo, then hands you the checkout.
  • School emails. In the U.S., forward them to [email protected] and Gemini turns them into calendar events and flags conflicts.
  • Texting Gemini. You can start a task over RCS, which needs a U.S. number on an eligible carrier. On Android and iOS it is under Settings > Channels > “Text with Gemini”.
  • More than 60 connected apps.

Google says the more advanced features start in the U.S. first.

How Google keeps it in check

Each task runs in its own hardware-isolated, sandboxed virtual machine on Google Cloud’s Agent Compute Engine, dedicated to one account. Prompt-injection filters and Google Safe Browsing limit which links Gemini will open. Google says “Gemini is designed to ask for your confirmation or pre-approval before clicking “Buy,”” and you can choose “Always allow” for routine steps, jump in while a task runs, disconnect apps or delete your chat history. Google also admits tasks can fail, and that some websites hand the final click back to you. Security researchers can report problems through its AI Vulnerability Reward Program.

What it means for developers

  • Your site may get agent visitors. Tasks signs in, fills carts and checks out on real websites. Clear forms, labelled buttons and checkout flows that don’t depend on hover tricks make it more likely an agent finishes the job, and a final confirmation step on your side is still yours to keep.
  • There is a way in for apps. Developers can apply to Google’s Connected Apps Platform to have their app work with Gemini, and Google says it plans to bring the Universal Commerce Protocol (UCP) into Gemini.
  • Payments go through AP2. If you run a store, the agent pays with digital tokens on supported cards through the Agent Payments Protocol, so watch your payment provider’s support for it.
  • The model is one you can call. Tasks runs on Gemini 4 Hydrogen, which is also in the Gemini API and AI Studio, so you can test how the same model handles your own agent flows. Building Android apps with Gemini is covered in our post on AI Studio’s full-stack Android apps.

How it compares

The change is mostly about price and reach. Per TechCrunch, the agent was limited to testers and then to U.S. Google AI Ultra subscribers; it now comes free in six countries on the model that also powers Google Search. Google’s post does not say how many tasks a free account can run, or what paid plans get on top, so limits for free users are not known yet.

Sources