Google's "Auto Browse" AI agent is designed to shop for clothes, plan trips, and buy tickets on behalf of users. But how well does it really work? The author, a tech enthusiast, decided to test the tool after being promised an "effortless" experience.
The first step was to start Auto Browse in Chrome. However, the author quickly realized that he had to log out of his Google account and refresh the browser to gain access to the feature. This seemed like a convoluted process, considering the promise of ease.
Once the author finally got into the tool, he began to test its capabilities with three different prompts: booking two tickets to a symphony concert, finding a leather jacket on Depop, and looking for two-person tent camping spots in April or May.
While the first prompt was resolved relatively quickly by Auto Browse, the outcome turned out to be incorrect - the seats were not side-by-side as requested.
The second prompt fared slightly better, with Auto Browse able to find a suitable jacket after some time. However, it didn't offer any nuanced reasoning for its choices, instead simply listing three options that had been top search results.
For the third and final test, the author was asked to book a camping trip in a specific area within three hours of San Francisco. After an extended period of time, Auto Browse produced less-than-ideal results, suggesting he visit another website for more information on availability.
The overall experience left the author underwhelmed with the tool's performance. He questioned how far it could be trusted and whether it was truly a worthwhile substitute for human interaction online.
The first step was to start Auto Browse in Chrome. However, the author quickly realized that he had to log out of his Google account and refresh the browser to gain access to the feature. This seemed like a convoluted process, considering the promise of ease.
Once the author finally got into the tool, he began to test its capabilities with three different prompts: booking two tickets to a symphony concert, finding a leather jacket on Depop, and looking for two-person tent camping spots in April or May.
While the first prompt was resolved relatively quickly by Auto Browse, the outcome turned out to be incorrect - the seats were not side-by-side as requested.
The second prompt fared slightly better, with Auto Browse able to find a suitable jacket after some time. However, it didn't offer any nuanced reasoning for its choices, instead simply listing three options that had been top search results.
For the third and final test, the author was asked to book a camping trip in a specific area within three hours of San Francisco. After an extended period of time, Auto Browse produced less-than-ideal results, suggesting he visit another website for more information on availability.
The overall experience left the author underwhelmed with the tool's performance. He questioned how far it could be trusted and whether it was truly a worthwhile substitute for human interaction online.