Google is working on an AI system called Project Jarvis that is meant to operate inside Chrome and complete ordinary web tasks for users. According to a report from The Information, the assistant is being built to control the browser directly rather than simply answer questions.
The idea is straightforward but significant: instead of asking an AI model for instructions, a user could have the system take action in the browser. That could make Project Jarvis less like a chatbot and more like a web task assistant.
What Project Jarvis is designed to do
Project Jarvis is being developed to perform everyday online actions without user intervention. The source article lists tasks such as searching the web, making purchases, and booking flights.
Those examples matter because they involve multiple steps. A person usually has to read pages, compare options, enter information, click buttons, and move between screens. Project Jarvis is meant to handle that kind of browser work directly in Chrome.
According to three people familiar with the project, the system works by taking regular screenshots of the browser window. It then analyzes what appears on the screen and performs actions such as clicking or entering text.
That means the assistant is not described as a separate website or a general-purpose operating system tool. Its role is tied to Chrome browser control, which gives the project a narrower focus than some other approaches to AI automation.
Why Chrome control is the key distinction
The source compares Google’s approach with a recent computer automation system from Anthropic. The difference is scope. Anthropic’s system is described as broader program access, while Project Jarvis is focused solely on controlling Chrome.
That narrower scope could shape how people use it. Chrome is where many common web tasks already happen, so an assistant that can read the browser window and act inside it could be aimed at routine consumer activity rather than specialized work.
The source says Google is targeting average consumers rather than developers and office workers. That positioning is important. It suggests Project Jarvis is not being framed mainly as a coding assistant or an enterprise productivity tool, but as a helper for ordinary web browsing tasks.
In practical terms, the browser becomes the workspace. If the assistant can search, click, type, and move through pages, the user may not need to guide every small step. The value would come from reducing the manual work involved in routine online actions.
The limits Google still has to solve
Project Jarvis is not presented as finished. The source article describes two major issues: speed and trust.
Sources say the AI needs several seconds to "think" before each action. That could make the system feel relatively slow, especially when a task requires many clicks or text entries. A few seconds may not sound like much in isolation, but repeated delays can add up during a multi-step web session.
The second issue is more sensitive. Google needs to address user concerns about sharing information such as passwords and credit card information with the system. That concern follows logically from the tasks Project Jarvis is meant to perform. Purchases and bookings often require private details, and a browser-controlling assistant may need to interact with pages where that information appears.
These concerns do not mean the system cannot work. They do show why browser automation is different from generating text. Once an AI assistant can take actions, users have to consider what it sees, what it enters, and how much control they are comfortable giving it.
How Jarvis fits into Google’s AI plans
Google plans to announce Project Jarvis alongside its new Gemini language model this December, according to the source article, but that timing is not set in stone.
The December timeline matches earlier reporting from The Verge about Google’s plans to launch its Gemini language model. However, that report indicated Gemini might not deliver major performance improvements over existing AI systems.
The source article connects this to a broader shift in AI. It says AI companies appear to be reaching a plateau in raw language model capabilities despite continued efficiency gains. In response, they are looking for new ways to make AI systems more useful.
Project Jarvis fits that pattern. Instead of focusing only on stronger language output, it points toward practical AI applications that can act on behalf of users. A system that controls Chrome is not just producing answers; it is attempting to complete tasks.
The earlier debate around the Jarvis idea
The name "Jarvis" had already appeared in discussions about Google’s AI strategy. Former Google UX strategist Scott Jenson criticized the company in May, saying it aimed to create a Jarvis-like assistant to keep users within Google’s ecosystem, motivated by fears that competitors might do it first.
That criticism frames Project Jarvis as more than a technical experiment. If an assistant becomes the layer through which people search, shop, book, and browse, it could influence where users spend time and which services they rely on.
For now, the confirmed details from the source remain limited. Google is developing Project Jarvis, it is meant to control Chrome, it uses browser screenshots to understand what to do, and it is aimed at common consumer web tasks. The open questions are whether it can act quickly enough, whether users will trust it with sensitive data, and whether Google will announce it this December as planned.