The Gemini Skills: the jump from assistant to autonomous agent inside Chrome

Google is preparing a new function that makes Gemini more than just a passive assistant within Chrome: it is testing what is internally called "Skills," a set of capabilities that would allow the model to run tasks automatically within the browser.

Today, in the versions already circulating in the United States, Gemini acts as a contextual support: you open it when you are on a page and ask him to clarify a paragraph, summarize an article or compare information between several tabs. That ability to "read" several tabs and condemn data - for example, take the relevant details of three open tabs on flights, hotels and activities and turn them into a unique plan - is precisely the basis on which Google wants to build more autonomous functions.

The concrete track came when an investigator published a capture that shows a new internal page, chrome: / / skills where you could define a "skill" with a name and instructions. This finding was disseminated on networks by Leo and can be seen in the publicly shared link itself: capture of discovery. The existence of that page suggests that Google is putting in place a system for Gemini to run concrete actions within the browser, not just answer questions.

According to what the company itself has explained in its presentations on Gemini and what has been seen in tests, the idea is to evolve from an assistant who explains and sums up to an agent who can do tasks: to search pages that you visited before with questions in natural language, to interact more closely with Google applications such as Calendar, YouTube or Maps, and presumably automate repetitive steps without having to change the tab constantly. For official context about Gemini and Google's strategy at IA it is appropriate to review your blog entries about the project: official introduction to Gemini.

Gemini's acquisition of executable "skills" within the browser opens up clear productivity possibilities: imagine delegating the task of comparing prices, extracting event schedules or filling out forms, and receiving a plan response or even a concrete action. However, this step also raises technical and ethical issues that should not be underestimated. An agent who can act on websites has to deal with permits, clear limits, transparency about what he does and why, and safeguards to avoid unwanted actions or abuses by third parties.

In the browser ecosystem there are already precedents for experimental features accessible through internal pages or lags, and the Chromium project offers the technical background that allows Google to test changes in Chrome before launching them in general: Chromium project site. But turning these internal tests into a public function will involve designing privacy controls, security audits and probably options for the user to limit what "Skills" can run and in what contexts.

The risks that are often noted by experts when an IA gains autonomy on the web include involuntary actions (clicks, forms), unnoticed filtering of sensitive data and the possibility that extensions or third parties will configure malicious skills if there are no strict controls. It is therefore reasonable to expect Google to implement explicit confirmations, granular permissions and an activity record for any automated operation, although there are still no public details on these measures.

For the moment the novelty is in internal tests and has no known launch date. In parallel, Google has made clear its intention to integrate Gemini more deeply with its own services and to enhance its ability to "act" in web environments, a movement consistent with the general tendency to bring language models to real and repetitive tasks. To follow the evolution of these tests, the discussions on security and privacy should be attentive to the official Chrome channels and technical reports that are emerging in specialized media.

If you are interested in testing early Chrome functions or following the experiments, the Chrome blog page usually publishes news and ads when the tests move to more open phases: official Chrome blog. And if what is concerned is the security implications, the documentation and discussion of the Chromium project provides a useful technical vision of how these capabilities are designed and tested.

In short, "Skills" for Gemini aim to transform the assistant into an agent capable of running tasks in Chrome. It is an advance with potentially huge impact on productivity, but also requires robust controls to protect data and avoid dangerous automations. Between public announcement and actual implementation there will be critical design decisions that will determine whether this function ends up being a discreet and secure help or a source of more privacy problems in daily navigation.