Gemini Tasks: How the Free AI Agent Mode Works and Where It Stops
Gemini Tasks turns the Gemini app from a question-answering chatbot into an agent that works through jobs for you. On November 17, 2026, Google announced the mode is free for users in the U.S., Japan, Korea, Canada, Aust
Gemini Tasks turns the Gemini app from a question-answering chatbot into an agent that works through jobs for you. On November 17, 2026, Google announced the mode is free for users in the U.S., Japan, Korea, Canada, Australia and India. It was previously available to select testers and AI Ultra subscribers under the name Spark. For developers and tech-curious readers, it is a useful look at how consumer AI agents are being packaged, and what guardrails come with them.
What is Gemini Tasks?
Chat gives answers; Task mode gives outcomes. Per Google's announcement, you describe a job and Gemini carries it out in the background: scheduling appointments, analyzing bills, shopping and managing lists. TechCrunch and The Verge both report that it runs on Google's new Gemini 4 Hydrogen model and replaces the Spark branding.
How to enable it
Open the Gemini app.
Flip the home-screen toggle from Chat to Task.
Describe the job.
In the U.S., Google also lists texting and email channels. To set up texting: Settings → Channels → Set up under Text with Gemini → verify your phone number.
What it can do (Google's descriptions)
Calls and bookings: calls local businesses and finds a slot in your calendar. Google says Gemini discloses it is an AI and businesses can opt out.
Bill analysis: reads past statements in email to explain a spike.
Shopping: compares products and checks out with Google Pay using the Agent Payments Protocol (AP2), with an optional spending cap.
Groceries: U.S. users can text items for an Instacart order.
Email to calendar: forwarded emails can become calendar entries.
The Verge adds that U.S. users get business calls to compare prices and check stock, plus sign-ins for sites that support Google Password Manager. Google notes calling is unavailable in Indiana, Louisiana, Minnesota, Montana and Nebraska.
The Gemini Tasks security model
The interesting part for a technical audience is the isolation design. Google says each task runs in a hardware-isolated virtual machine dedicated to your account, with prompt-injection filters on search results and web pages, and Safe Browsing limiting which links Gemini opens. These are vendor claims; the sources here do not include independent testing.
Human-in-the-loop controls are explicit:
Gemini is designed to ask for confirmation before critical steps such as buying.
Routine actions can be set to "Always allow".
You can adjust task settings, disconnect connected apps and delete chat history.
Google says mistakes can still happen, and some sites will hand the screen back for the final click.
Practical rules for agent tools
- Start with low-stakes tasks (lists, calendar cleanup)
- Keep purchase confirmations on until results are proven
- Set spending caps well below card limits
- Verify the outcome (calendar entry, order) yourself
- Disconnect apps you do not need; secure the Google account with 2-step verification
My take
I have not tried Gemini Tasks yet — it only went free yesterday, and I have not seen the toggle appear for me. My first job for it would be a low-stakes one, probably a grocery list or a calendar clean-up, because I want to see how it handles something simple before I let it near a payment or a phone call. The part I like is that it is designed to pause and ask before it buys anything; the part I would watch is how much access it needs to do its job well. For now, I would keep confirmations on and set the spending cap low, and check every result myself until it earns more trust.
Chat vs Task: choosing the mode
Use Chat when you want an answer and Task when you want an outcome. A question about fair pricing for a car service is Chat; booking the appointment is a Task. Because Task mode can touch your calendar, email and payments, keep everything exploratory in Chat until you trust how the agent behaves.
Where it is available
Per Google, the free rollout covers the U.S., Japan, Korea, Canada, Australia and India, while the advanced features start in the U.S. The sources above give no date for other regions, so readers elsewhere may not see the toggle yet.
Where this fits
The Verge notes Tasks is one of several agentic features that have emerged over the past year, including Claude Duet, OpenAI's Dots and Meta's Muse. The common design question is the same: how much can the agent do before it must stop and ask?
Your turn
Which chore would you delegate to an agent first, and which would you never let it touch? Tell us in the comments.
Sources
Put Gemini to work for you (Google, Nov 17, 2026; vendor announcement)
Google's Spark AI agent becomes 'Tasks' — and now it's free (The Verge, Nov 17, 2026)
Cover photo: Organized Workspace with Planner and Smartphone (Pexels, free to use)
Originally published by Dev.to Security. Aggregated on AIWithGhost for educational purposes — full credit and traffic to the original publisher.