Google has launched Gemini Spark, an AI agent that carries out real web tasks inside the Chrome browser. It goes beyond answering questions to directly handle repetitive web tasks such as shopping, reservations and document completion, a strategy aimed at expanding Chrome’s role with “action-oriented AI.”

Tech outlet TechRadar reported on Aug. 5 that Google unveiled Gemini Spark, an AI agent linked to the Chrome browser. The feature can explore websites on a user’s behalf and perform tasks such as price comparisons, reservations and form-filling in the background.

The biggest difference from existing Gemini, which focused on information search and answers, is that Gemini Spark directly operates the browser. Users can instruct tasks through the “Ask Gemini” panel at the top of Chrome. Spark first presents a work plan, then asks for permission to connect to the browser. After approval, it moves across multiple web pages to carry out the task. It is also designed to continue working in the background even if the user closes the Chrome window after the task begins.

The requirements are somewhat strict. It needs the latest version of Chrome on Windows 11 or macOS and a personal Google account. Users must also have a Google AI Pro or Google AI Ultra subscription and set Chrome Safe Browsing to Standard protection or Enhanced protection mode. It can then be used by enabling “Let Gemini browse for you” in Chrome AI settings.

The core of Gemini Spark is automating repetitive web tasks. For example, if a user asks it to compare 65-inch TVs on Amazon, Best Buy, Costco and Walmart and find the lowest price after applying discount codes, Spark visits multiple retailers to compare prices and promotions. It then adds the item to the cart and prepares the purchase up to the final step before payment.

It does not complete payments automatically. At stages involving money, it must request user approval and then stop the task.

It also supports bookings. If asked to plan a family outing for two adults and two children, it creates an itinerary considering operating hours and travel routes, and proceeds with restaurant reservations and museum ticket purchases up to the final approval step.

Document completion is also possible. For tasks such as applying for a library membership card, it automatically fills out an application using saved address and contact details, but is designed so the user presses the final submit button. Google sees Spark as particularly strong at document tasks that use information from a user’s Google account.

Google’s policy is to secure both automation and safety through such approval procedures. For important tasks such as purchases or entering sensitive personal information, it ensures the AI does not act unilaterally and requires user confirmation. TechRadar said this approach shows a “balance between useful automation and common-sense safety guardrails.”

Some in the industry also say Gemini Spark marks a turning point in Google’s AI strategy. While existing generative AI focused on answering questions or explaining how to use something, Gemini Spark has evolved into “agentic AI” that carries out work by directly operating the browser.

In particular, by directly integrating an AI agent into Chrome, analysts say Google’s strategy has become clearer to reshape everyday browser tasks such as search, shopping, reservations and document completion around AI. For now, usage is limited at the outset because it requires the latest operating system and Chrome, a paid AI subscription and security settings.

Even so, Gemini Spark is drawing attention as an example showing web browsers evolving beyond simple internet access tools into AI platforms that carry out real work.