Google is gradually turning Gemini into an everyday agent that does more than answer questions: it can prepare purchases, call businesses, and interpret data from connected devices. The new experiments affect people who want to spend less time organizing routine tasks, as well as companies using AI for customer interactions. Many of these capabilities are still tests, however, and access may depend on your country, device, or subscription.
From assistant to everyday agent
Artificial Intelligence (AI) describes systems that recognize or generate content and use data to support decisions. An AI agent goes further by taking actions for you within defined limits. With Gemini, that can now mean moving from a product search to checkout or placing a call to a business.
This development changes how you interact with online services. Instead of navigating several websites, apps, and phone menus yourself, you state a goal and let AI complete some of the steps. That saves effort, but it can also make it harder to see how Gemini selects offers, which information it shares, and when an action is truly complete.
The most prominent new trial involves online shopping. According to a report on Google’s Flipkart test, selected shoppers in India can buy certain smartphones, electronics, and mobile accessories directly through Gemini or Google’s AI Mode. A Buy button on some listings launches a Flipkart-branded checkout flow within the AI interface.
The trial is small and does not cover every user in India or Flipkart’s complete catalog. Google reportedly planned a wider rollout later in October, but the company only confirmed that it regularly tests new features and experiences. The technology powering this particular trial was not disclosed.
Shopping and calling mark the next stage
In so-called agentic shopping, AI handles parts of a transaction instead of merely recommending products. Google previously introduced the Universal Commerce Protocol (UCP), an open standard providing common rules for communication between AI services and retailers. UCP is intended to support both Google-hosted checkout and transfers to a retailer’s website, but it is unclear whether the Flipkart test uses it.
For you, this could shorten the path between wanting and buying a product. You might search for a suitable smartphone, choose one of the listed models, and proceed directly to Flipkart’s checkout. The test also raises a potential conflict of interest: Google is Flipkart’s technology partner and invested about $350 million in the company in 2024, receiving a minority stake.
Gemini is also being tested on the less appealing parts of making phone calls. With Google’s “Call for Me” feature, AI calls a business on your behalf, navigates voice menus, and waits on hold. Examples in the source include making reservations, rescheduling an appointment, and asking whether a product is in stock.
You can follow the conversation through a live transcript, meaning a written record produced as the call happens, and take over at any point. Gemini uses your own number and can share personal information that you have approved in advance. Google says it is starting on a small scale because real conversations contain nuances that do not always fit predictable workflows.
Access is highly restricted. Initially, only US owners of a Pixel 11 with a paid Gemini subscription can use the feature, and they also need the beta version of Google’s Phone app. The source does not provide a specific subscription price, and availability in Switzerland has not been confirmed.
Avatars, health, and the home expand Gemini’s role
Gemini is gaining a visible persona as well as more ways to act. Gemini 3.8 Live with Live Avatar lets you talk to an animated character whose lip movements and facial expressions are designed to match its responses. The avatar can also display relevant information on screen while it speaks.
Google says the system can move between 97 languages without reduced video quality or visible drift in the animation. One demonstration switched between English and Japanese while keeping the mouth movements aligned with the speech. These performance claims come from the provider and were not independently verified in the supplied sources.
Live Avatar is currently limited to Gemini Enterprise customers. Businesses can select preset characters or create custom avatars. Google says the output includes its invisible SynthID watermark, a technical marker for AI-generated content, along with safeguards intended to respect people’s identities.
Health data is becoming part of this broader assistant model as well. The Health Guardian for Pixel Watch and Fitbit Air analyzes several weeks of sensor readings and is intended to show trends related to blood pressure, insulin resistance, and breathing quality during sleep. It is designed for third- through fifth-generation Pixel Watches and the Fitbit Air band.
The insulin-resistance analysis is intended to detect gradual metabolic changes without a blood sample. Google draws clear limits around that claim: the feature does not diagnose disease, is not a preliminary diabetes test, and does not monitor blood sugar in real time. It can present information that sounds medical, but that information is not equivalent to a clinical diagnosis.
In the connected home, AI performs another type of filtering. Modern security cameras can move beyond generic motion alerts by using visual language models, which combine image recognition with written summaries. A comparison of Apple, Amazon, and Google Nest illustrates why an alert such as “small dog chasing a chicken” can be more useful than a broad “motion detected” notification.
A text description may communicate the essential event faster than a video feed that first needs to load. Google Nest is among the providers offering these descriptions, alongside Ring, Blink, Wyze, and Arlo. The comparison also reports that Apple charges up to three times as much as competing services for the feature, although the source does not provide a specific Google price.
Pros and Cons of an everyday agent
Pros:
- Less routine work – Gemini can handle hold queues, voice menus, and parts of a checkout process for you.
- Faster interpretation – Camera descriptions and health trends can condense many signals into understandable updates.
- More ways to interact – Live transcripts, spoken responses, and multilingual avatars can support different use cases.
- Visible control points – During an AI-handled call, you can monitor the transcript and take over whenever necessary.
Cons:
- Restricted access – The announced capabilities may be limited to India, the US, enterprise customers, or particular Google devices.
- More sensitive data – Purchases, phone numbers, approved personal details, camera footage, and health readings all reach into private life.
- Risk of mistakes – AI may misunderstand conversational nuance, a scene captured by a camera, or a trend in body data.
- Opaque selection – Financial and technical partnerships may affect which retailers or offers appear in an AI interface.
What this means for you
For beginners: Start with a task whose consequences are easy to review. A camera summary or a monitored business call is easier to supervise than an automated purchase or the interpretation of a health reading. Before using a feature, check whether you can see each stage, stop it, or take over, and review which details Gemini is allowed to share with another party.
For advanced users: You can get more value by setting different limits for different types of work. For example, Gemini could gather product information while you retain control over the final selection and payment. With Call for Me, the live transcript can serve as a checkpoint, while health and camera notifications should be compared with the underlying readings or recordings.
The picture remains incomplete for Switzerland. Shopping is being tested in India, Call for Me starts under narrow conditions in the US, and Live Avatar targets enterprise customers; none of the sources confirms general availability in Switzerland. Support for 97 avatar languages also does not establish how well the service handles Swiss Standard German or other languages used in the country.
Gemini is credibly moving from an answer service toward an interface for real-world actions. Its main benefit is reduced routine work and faster interpretation of information, while broad availability and many prices remain unknown. The unresolved risk is that convenient automation can obscure decisions, data flows, and mistakes, especially when purchases, private calls, health readings, and footage from your home are involved.
Sources
- Google tests buying from Walmart-owned Flipkart through Gemini and AI Mode in India – TechCrunch, 2026-09-27
- Google testet “Call for Me”: Gemini ruft für Nutzer bei Unternehmen an – The Decoder, 2026-09-25
- Gemini 3.8 Live with Live Avatar gives Google’s AI a face – The Verge, 2026-09-24
- Google Pixel Watch: Wie die Uhr künftig vor Diabetes-Anzeichen warnen kann – t3n, 2026-09-25
- Can Apple Home’s AI camera features outsmart Amazon’s and Google’s? I put them to the test – The Verge, 2026-09-25
Image: Caio via Pexels