Google Builds Gemini Into Chrome and Teaches It to Browse

Google has begun rolling out a full Gemini integration into Chrome for Android, including an "auto browse" mode that can scroll, tap, type and complete multi-step tasks on a user's behalf. The announcement landed during the company's Android Show on 12 May and confirms what many had assumed since the desktop version of Gemini in Chrome started shipping in January: Google plans to make the browser itself agentic.

Tapping the new Gemini button next to the address bar opens a bottom sheet that can answer questions about the current page, summarise long articles, draft replies in Gmail and pull recipe ingredients into Keep. The more interesting feature, and the one that has drawn most of the attention, is auto browse. Given a goal in natural language, Chrome will navigate to the relevant sites and operate them autonomously until the task is done, pausing for confirmation before anything sensitive such as a payment.

What auto browse actually does

The example Google demonstrated involved finding parking near a venue. The user pastes an event ticket into Gemini, asks for parking nearby, and Chrome opens SpotHero, fills the dates and times from the ticket, and presents the booking page for the user to confirm. 9to5Google reports the feature is built on Gemini 3.1 and will be available only to AI Pro and AI Ultra subscribers, on select Android devices with at least 4 GB of RAM, beginning at the end of June in the United States.

The cautious rollout matters. Browser-based agents have a poor track record so far. They are easily tricked by prompt injection from page content, and they routinely click the wrong things on sites with cluttered layouts. Google's confirmation gate on sensitive actions, payments, account changes, message sends, is an admission that the model is not yet trustworthy enough to operate without supervision.

The browser as battleground

Microsoft already has Copilot deeply embedded in Edge. Perplexity is shipping Comet. The Browser Company's Arc is rebuilding itself around an AI agent. Chrome controlling roughly two-thirds of the browser market gives Google an obvious advantage in distribution, but the company has been notably slower than its rivals to ship agentic features, partly because of the regulatory scrutiny already focused on its search business. The May announcement looks like Google deciding the wait is over.

For users, the question is whether browsing-as-a-prompt actually saves time. Telling a model to "find me parking" still requires the user to verify the result. If the verification step is slower than just booking the parking yourself, the agent is theatre rather than tool. We looked at the broader question in our recent piece on enterprise agent deployments, where similar tensions are showing up at scale.