Free Intensive Nov 21–22Free 2-day hands-on Vibecoding Intensive on Nov 21–22 at 16:00 MSK (Sat–Sun) · Live Cohort 3 starts Nov 23Join Free →Cohort 3 (Nov 23)
ReleaseInsight · On RadarGoogle AI Blog→ original

Gemini in Task Mode Calls Auto Shops, Pays for Purchases and Parses Bills

Google has turned Gemini from a chatbot into a doer. In Task mode, it calls auto repair shops on its own, pays for purchases and parses bills from email. Each task runs in a separate isolated virtual machine on the new Gemini 4 Hydrogen model. The basic features are free for hundreds of millions of people in six countries, while the expanded features are so far available only in the US. For developers, this is a signal that agentic payments and calls are moving into the mass market.

AI-processed from Google AI Blog; edited by Hamidun News
Gemini in Task Mode Calls Auto Shops, Pays for Purchases and Parses Bills
Source: Google AI Blog. Collage: Hamidun News.
Listen to article
Article Fact Sheet & Key Figures
ParameterDetails
Primary sourceGoogle AI Blog ↗
Editor-in-Chief
Key metric #1Payments go through the AP2 protocol with Visa and Mastercard tokens and a spending limit
Key metric #2Gemini connects to more than 60 apps, including Instacart
Key metric #3Before the "Buy" button, the assistant asks the user for confirmation

Chatbots have long been good at giving advice, but they rarely get things done for you: calling a service center, waiting on hold, finding an open slot and booking it. Google has decided to close that gap. On November 17, the company announced that Gemini now carries out everyday tasks end to end: it makes calls, pays, looks for discounts and organizes things on your calendar. For those building agents and payment services, this is one of the clearest examples of agents launching at mass scale.

What Gemini Can Do on Its Own

The Gemini app now has a Chat/Task toggle on its home screen. In Task mode, the assistant takes on a job and works on it in the background. Josh Woodward, Vice President of Google Labs, demonstrated this with his own examples.

In the first example, it's time for service on a 2019 Honda Odyssey. Gemini knows the car and the address from Personal Intelligence, finds the two nearest dealerships on Google Maps and calls both at the same time. It waits on hold in place of the owner, chooses the service center that allows advance booking and puts an early Friday slot on the calendar. The whole thing took under five minutes. At the start of the call, the assistant states that it is an AI, and businesses can decline such calls.

In the second example, there is a $389 electricity bill. Woodward gave Gemini access to the utility's account portal through the Google Password Manager integration and asked it to go through past bills in his email. A few minutes later, the assistant found the cause: the heat pump was switching on around 4 p.m., during the peak tariff window.

In the third example, Gemini compared USB-C chargers through Shopping Graph, confirmed wattage and color, placed the order with Google Pay and found a $10 promo code. Groceries can be added to Instacart by SMS, and a photo of an empty cereal box works as an "add this" command. One more feature: emails forwarded from a school to [email protected] turn into calendar events.

How Payments and Protection Work

Purchases run through the Agent Payments Protocol (AP2). Google works with Mastercard and Visa, and American Express is promised later. Payments are processed with digital tokens, and points, cash back and card purchase protection are preserved. Before paying, you can set a hard spending limit, and receipts land in Google Wallet, which from today is also available on iOS. Google plans to add the Universal Commerce Protocol (UCP) next to make the path from choice to payment faster.

Each task runs in an Agent Compute Engine virtual machine in Google Cloud. The machine is dedicated to one account, hardware-isolated and not shared with other users. This is only one layer of protection. There are also filters against prompt injection in search results and web pages, and Google Safe Browsing limits which links can be opened. Before critical steps, such as the "Buy" button, Gemini asks for confirmation. For trusted actions, such as adding cereal to the cart, an "Always allow" mode can be turned on. Google is also inviting researchers to find vulnerabilities in agentic systems through its AI VRP program.

Where the Limits Are and What Developers Should Do

The update runs on the new Gemini 4 Hydrogen model. Basic features are free in the US, Japan, Korea, Canada, Australia and India. Expanded features are currently available only in the US. Calls do not work in Indiana, Louisiana, Minnesota, Montana or Nebraska, and the SMS channel via RCS requires a US number with a compatible carrier.

Google openly admits that errors are not yet solved. On some sites, the agent completes nearly the whole path and then hands the screen back so the user can press the final button. The web has blocked bots by default for decades, because most of them were malicious. Now the company promises to work with major sites and security providers on open, verifiable standards.

Gemini is already connected to more than 60 apps, and developers can apply to the Connected Apps Platform. For product managers, the takeaway is simple: if your service takes bookings, orders or payments, it will be used by more than just people. It's worth checking how your site behaves with an agent: does it offer tokenized payment, clear confirmations, an API instead of a CAPTCHA? Those who prepare first will get a stream of orders from assistants. Those who block every bot in sight risk losing these customers.

A separate question is trust in the ecosystem. Google is openly selling the idea that "your data is already with us, so there's no need to send it anywhere." That is convenient, but it locks in a link: the more mail, passwords and cards you keep in Google, the more useful Gemini becomes. Competitors that don't own that kind of data will have to rely on open protocols such as AP2 and UCP.

Story

Key Questions Answered

What Gemini Can Do on Its Own?

The Gemini app now has a Chat/Task toggle on its home screen. In Task mode, the assistant takes on a job and works on it in the background. Josh Woodward, Vice President of Google Labs, demonstrated this with his own examples. In the first example, it's time for service on a 2019 Honda Odyssey. Gemini knows the car and the address from Personal Intelligence, finds the two nearest dealerships on Google Maps and calls both at the same time.

How Payments and Protection Work?

Purchases run through the Agent Payments Protocol (AP2). Google works with Mastercard and Visa, and American Express is promised later. Payments are processed with digital tokens, and points, cash back and card purchase protection are preserved. Before paying, you can set a hard spending limit, and receipts land in Google Wallet, which from today is also available on iOS. Google plans to add the Universal Commerce Protocol (UCP) next to make the path from choice to payment faster.

Where the Limits Are and What Developers Should Do?

The update runs on the new Gemini 4 Hydrogen model. Basic features are free in the US, Japan, Korea, Canada, Australia and India. Expanded features are currently available only in the US. Calls do not work in Indiana, Louisiana, Minnesota, Montana or Nebraska, and the SMS channel via RCS requires a US number with a compatible carrier. Google openly admits that errors are not yet solved. On some sites, the agent completes nearly the whole path and then hands the screen back so the user can press the final button.

ZK
Zhemal Khamidun — CPO at Alpina Digital, product manager of AlpinaGPT. Building AI products since 2019. Hamidun News is a pet project that grew into a production aggregator with 6,000+ readers.
✦ Unified AI Workspace · Nomad

Test Frontier LLMs & Autonomous Agents in One Workspace

Run Claude Opus 5.5, GPT-6, Gemini 3.1, and DeepSeek side-by-side in Nomad AI (nomadnet.ai) — unified model routing, deep research, and multimodal workflows under one subscription.

Try Nomad AI →Launch Nomad Agent
What do you think?
Loading comments…