AI & Society

Gemini Tasks Goes Free, but It Still Needs Your Permission

British newspaper cartoon showing a cheerful Google Gemini AI robot explaining task automation to a curious man, with a laptop displaying permissions for email, calendar, files and location.
Gemini is ready to take on everyday tasks without a subscription, but users still control which services and information the AI assistant can access.

Google is expanding Gemini Tasks, its feature for handing off everyday jobs, to users who do not pay for a premium AI subscription. The company says the new capabilities are rolling out in the United States, Japan, South Korea, Canada, Australia and India. That distinction matters: someone reading about the feature from Britain or continental Europe should not assume the same free access is already available in their country.

The 17 November announcement describes a more practical form of AI assistance. Rather than simply suggesting when to book a car service or how to assemble a shopping list, the assistant can use connected services to take steps in the process. Google gives an example of calling dealerships, checking the user’s calendar and arranging an appointment. It says the task-running system discloses that it is AI when calling a business.

What Gemini Tasks changes for customers

A chatbot normally returns text that a person must act upon. A task agent has access to tools, so it can move part of the work forward without waiting for every individual instruction. That change can reduce friction when a chore spans email, a calendar, a website and a telephone call. It can also make the consequences of a misunderstanding more serious.

Imagine asking for a replacement household appliance. An ordinary assistant might compile a shortlist. A task agent could begin checking prices and availability. It might compare models, fill a basket and prepare checkout, depending on the supported functions and permissions. Every extra capability adds value but also introduces decisions about spending limits, privacy and the point at which a human must confirm the choice.

Google says users can switch from Chat to Task within Gemini, and that people in supported regions can begin requests through mobile entry points and forwarded emails. The details of availability and advanced features vary, so a user’s app remains the best guide to what is actually enabled on their account. A regional announcement is not evidence of universal access.

How Google says it separates the agent from your accounts

The assistant needs to read information to organise a schedule or compare products. Google describes running tasks inside a hardware-isolated virtual machine dedicated to the user, with filters designed to detect malicious instructions embedded in web pages and other untrusted material. Those measures aim to reduce the opportunity for an outside document to redirect the assistant into taking an action unrelated to the user’s request.

Google’s more technical Agent Compute Engine explanation discusses isolated workspaces, short-lived credentials and auditability. For ordinary users, the significance is simpler: the system is not supposed to be given a permanent, unrestricted key to every connected service just because it has been asked to complete a small chore. The design can help limit risks, but it cannot eliminate mistakes.

The company also says Gemini pauses before critical actions and can ask for clarification if instructions are ambiguous. It may offer approval shortcuts for routine steps. That convenience needs careful judgement. Authorising the addition of groceries to a shopping basket is different from authorising a purchase, a recurring subscription or a message sent to a third party.

What businesses may experience

The change affects people on the other side of an automated task. Shops and service providers may receive more calls or online requests made by software rather than by a human customer. Good disclosure helps staff understand why a caller sounds unusual and whether it has authority to confirm a booking. Businesses may also need policies for handling requests from agents, particularly when the request could create a contractual or financial obligation.

Customers should not assume that an AI-arranged booking has been confirmed merely because an assistant reports success. A confirmation number, message from the provider or updated appointment record offers more dependable evidence. It is worth checking dates, names, amounts and cancellation terms in the same way as with a human booking.

The expansion also raises the same questions LiveAIWire has explored in its coverage of agentic shopping and consumer trust. When software acts on behalf of a customer, sellers still need to provide clear terms and customers still need to know what they agreed to buy. An assistant is a new interface, not a replacement for ordinary consumer protections.

A practical test for users

Start with a task whose outcome is easy to verify and whose failure is inexpensive. Ask the assistant to propose appointment slots, reconcile a calendar with a list of commitments or prepare a shopping comparison. Review the result before granting access to additional accounts. If the task is supposed to send a message, confirm the recipient and wording. If it involves payment, check the complete amount and whether it repeats.

It is also worth separating tasks that require judgement from tasks that only require patience. Software may be excellent at finding open appointment slots but less reliable at deciding which service is suitable. The human still needs to supply the preferences and constraints that make an outcome useful. A car service booking at the wrong branch is not much of a time-saver.

LiveAIWire has examined how payment approval systems are being built into AI tools. The same principle applies here: making tasks easier should not blur the moment at which a suggestion becomes an action.

The important news is not that a free chatbot has acquired another button. It is that a major technology company is trying to make task delegation a normal consumer habit. Whether it becomes genuinely useful will depend on dependable results, clear permissions and a user’s ability to correct mistakes before they matter.

About the Author

Stuart Kerr is Technology Correspondent at LiveAIWire, covering artificial intelligence, cybersecurity and the social impact of emerging technology. LiveAIWire is an independent, human-led technology publication using AI-assisted research, editorial production and original AI-assisted editorial illustrations under his direction.