When I finally got to experiment with auto-navigation (for real this time), I took Google’s digital task suggestions as a starting point and chose online tasks that might be useful in my own life.
Every time you interact with Generative AI tools, a healthy sense of skepticism – and caution – is essential. Google even includes a disclaimer built into its Gemini Chatbot reminding users that it makes mistakes. The automatic navigation tool goes even further. “Use Gemini with care and take control when necessary,” reads the persistent text that appears in the chatbot’s sidebar every time auto-navigation is running. “You are responsible for Gemini’s actions during tasks.”
Before you try it, you should also think about the security risks associated with this type of automation. Generative AI tools are susceptible to compromise by rapid injection attacks on malicious websites. These attacks attempt to distract the bot from its task. The potential vulnerabilities in Google’s auto-navigation haven’t been fully examined by outside researchers, but the risks may be similar to other AI tools that take control of your computer.
Additionally, be especially careful if you use automatic navigation to make purchases. Google has safeguards in place that flag certain actions, like purchasing products or posting to social media, as sensitive and requiring user approval to continue. Still, I didn’t know how the robot would behave and worried about the havoc it could potentially wreak on my computer. credit cardlet alone relaying financial information to him in the first place.
Here is the first invitation I sent him, card in hand:
I want to book two tickets to the SF Symphony tonight. I don’t want to pay for an orchestra seat, but the tickets don’t have to be the cheapest available. Please choose the two seats next to an aisle.
It’s a little weird to see Google’s AI agent clicking in the tab. First, I saw him using Gemini 3, Google’s latest model, to strategize and set goals, like getting two seats in the symphony aisle, in the sidebar text box for a few seconds. This process resembles a chatbot using a “reasoning” modeldiscussing the steps to take before moving forward. Then the clicking begins. Every step the bot takes as part of a task is recorded for users.
Auto Browse’s ability to multi-step tasks without getting distracted was significantly better than similar agent tools I tested last year. He went to the right website, chose the right performance, and clicked through several seating sections to gauge availability. Everything listed in the log seemed to match what he actually performed.
After a few minutes of working on the tickets for the symphony, the robot stopped clicking. I received a notification to take over and I press the Order Now button. At a glance, the AI tool had seemingly delivered what I asked for, and quite quickly.
But if I had without a doubt ordered the two seats Auto Browse chose for a date at the Symphony, the night would have most likely ended with my boyfriend forcing me to sleep on the couch.
