Ask your browser to do something for you and, for the first time in Chrome's long life, it might actually try.
Google has folded its latest Gemini 3 model deeper into Chrome — not as a novelty popup, but as a persistent sidekick and, with a new feature called auto browse, as an agentic helper that will roam websites and attempt to complete multi-step tasks on your behalf.
What just changed
Starting today for AI Pro and AI Ultra subscribers in the U.S., Chrome introduces a persistent Gemini sidebar that sits to the right of your tabs. It brings together three headline features: the multimodal Nano Banana image tool for on-the-fly image edits, an incoming Personal Intelligence layer that will tap your connected Google apps for contextual answers, and auto browse, an automation mode that can traverse websites, fill forms, and assemble results across pages.
Google says auto browse can do things like research flight and hotel costs across date ranges, filter apartment listings, gather quotes for a plumber, or even re-order items while hunting for coupon codes. If you let it, the agent can use Chrome's saved cards and Password Manager (with permission) so it can sign in where necessary — though final, sensitive actions typically require a human tap to complete.
The company is also backing the feature with an industry-facing standard called the Universal Commerce Protocol (UCP), developed alongside merchants and marketplaces so agents can act more reliably across stores.
Why Google is pushing this now
Browsers are once again the main battleground for AI. OpenAI's Atlas and a wave of AI-first browsers pushed the idea that the browsing experience could be built around an assistant. Chrome, which still dominates market share, is retrofitting those capabilities instead of being replaced. Integrating Gemini inside Chrome also lets Google leverage its own ecosystem — Gmail, Calendar, Maps and Drive — for more context-aware assistance.
You can read more about Google's broader mobile agentic plans in their AI mode rollout for Chrome on mobile, which expanded agentic booking and other controls earlier this month. And the deeper workspace hooks being added mirror efforts to let Gemini search inside a user's Google account for tailored answers, similar to the company's Gemini Deep Research work integrating Gmail, Drive and Chat into search workflows. For tasks tied to navigation or location, the Maps copilot experiments are another example of this push toward context-rich assistance with Gemini in Maps.
How it behaves — and where it stops
In demos, Auto Browse opens its own tab and performs what look like ghost clicks — clicking through pages, filling fields and reporting back the steps it took. Google emphasizes a few guardrails: the agent will pause and request explicit confirmation for purchases, posting to social media, or any other action classified as sensitive. In its blog announcement the company noted users remain in control and can disconnect apps or opt out of Personal Intelligence at any time.
That said, these systems are not magic. Reporters and industry observers point out that browser agents are finicky in real-world settings: websites change, flows break, and malicious pages can attempt to trick an automated agent with prompt injection attacks. Google says it has built new defenses into Gemini in Chrome, but it also describes auto browse as a cautious rollout — subscribers in the U.S. will get early access while the company gathers more feedback.
The trade-offs: convenience vs. risk
Auto Browse is the next step beyond autofill. Instead of just filling a card number, the agent might hunt for the cheapest travel dates or sift through your favorited apartments and drop the non-pet-friendly ones. That's a real time-saver when it works. But handing an AI the keys to click, sign in, and act introduces fresh privacy and security questions.
Threat models include malicious sites that can misdirect agents, mistakes in intent interpretation that lead to wrong purchases or form submissions, and surprise privacy leakage if connectors are promiscuously enabled. Chrome's approach — pausing on sensitive items and asking for confirmation — mitigates but does not eliminate those concerns. Wired's early testing notes that Google still frames you as responsible for the agent's actions, a point worth keeping in mind.
Where this fits in the bigger picture
For power users, the Gemini sidebar plus auto browse packs a productivity punch: imagine drafting an email that references flight options the agent found, or transforming a web image inline with Nano Banana without hopping between tabs. The promise is convenience; the reality will hinge on robustness.
TechCrunch and other outlets have pointed out that agents often stumble when faced with the messy diversity of the web. That means adoption will likely be gradual: curious subscribers will experiment, Google will iterate, and the feature will expand only as reliability and safety improve.
If you plan to try it, keep a few habits in mind: limit which apps and credentials you connect, watch auto browse sessions as they run, and be prepared to intervene on purchases or logins. If you use a MacBook or other desktops supported by the rollout, test in low-risk scenarios first.
Chrome's move is not just a product update. It signals how major platforms intend to make AI operational in everyday web browsing — which means browsers will increasingly be judged less on raw speed and more on how safely and accurately they let assistants act on your behalf. The next few months will tell whether agentic browsing becomes a convenience people trust, or another promising feature that needs more polish before it earns mainstream confidence.