When I was finally able to experiment with Auto Browse (for real this time), I took Google’s suggestions for digital chores as my starting point and chose online tasks that might be helpful in my own life.
A healthy sense of skepticism and caution is important whenever interacting with generative AI tools. Google also includes a disclaimer in its Gemini chatbot that reminds users that it makes mistakes. The Auto Browse tool goes a step further. “Use Gemini carefully and take control if necessary,” reads persistent text that appears in the chatbot sidebar every time Auto Browse runs. “You are responsible for Gemini’s actions during chores.”
Before you try it, you also need to think about the security risks associated with this type of automation. Generative AI tools are at risk of being compromised through rapid injection attacks on malicious websites. These attacks attempt to distract the bot from its task. Potential vulnerabilities in Google’s Auto Browse have not been fully investigated by outside researchers, but the risks may be similar to other AI tools that control your computer.
Also, be extra careful if you’re using Auto Browse to make a purchase. Google has security measures in place that consider certain activities, such as purchasing goods or posting on social media, as sensitive and require user approval to continue. Still, I was unsure how the bot would behave and I was worried about how much havoc it could potentially wreak on my credit cards, to say nothing of handing over financial information to it in the first place.
This is the first message I sent, card in hand:
I want to book two tickets to the SF Symphony tonight. I don’t want to pay for orchestra seating, but tickets don’t need to be the cheapest available. Please choose two aisle seats.
It’s a bit strange to see Google’s AI agent clicking around in tabs. At first, I saw it using Google’s latest model, Gemini 3, to strategize and define goals, like getting two aisle seats at Symphony, in a sidebar text box for a few seconds. This process looks similar to a chatbot using a “reasoning” model, talking about steps to take before moving forward. Then, the clicking starts. Every step taken by the bot as part of a task is logged for users.
Auto Browse’s ability to perform multistep tasks without getting lost was much better than similar agent tools I tested last year. It navigated to the right website, chose the right performance, and clicked through multiple seat sections to assess availability. Everything listed in the log appears to be as it was actually executed.
After working on Symphony tickets for a few minutes, the bot stopped clicking. I received a notification to take charge and press the order now button. At a glance, it appears that the AI tool delivered everything I asked for, and very quickly.
But if I had unquestioningly ordered two seats selected by auto browse for a date at the Symphony, the night would likely have ended with my boyfriend making me sleep on the couch.
<a href