The browser of most people The tabs are filled with unrelated press articles. Mine are filled with AI agents and ghost clicks.
I have four instances of the OpenAI Chatgpt agent – The generative AI tool published last week, which can do research and do tasks on the web – it is already open with each executed in its own tab. I gave these first four agents relatively simple jobs based on Chatgpt suggestions. One click to find a birthday present on the Target website, and another generates a pitch pitch on robotic dogs. I open a fifth tab in order to try something more experimental: I want to see how good this chatgpt agent is good at chess.
After typing certain instructions, I look at a ghostly cursor floats on my screen and the chatgpt agent goes to Chess.com and plays an online opponent, all in a virtual browser. Things go south fairly quickly. The strategy of the game is not what makes the AI tool tripping, it is the act of moving the chess parts which prove to be the most difficult. “I focus on a specific positioning while I continue to play despite bad clicks,” said the agent in his internal newspaper before finally leaving and letting me know that the controls were too difficult to navigate.
In recent years, browser developers have integrated AI tools with intermediate success. However, in recent weeks, the idea of a web browser improved by a generative oven chatbot has resurfaced with the release of the Openai chatgpt agent and the comet of Perplexity.
The two versions are very different in their execution. Comet is an autonomous browser, so you can use it to surf the web, then summon the AI assistant to help write an email or complete a subordinate chore. Openai built its navigation tool inside a chatbot; You speak to the chatbot via a web interface to give it tasks, then the bot runs its own virtual browser inside your browser to complete them.
The two versions can take control of the cursors, enter the text and click on the links. If this trend takes off, these types of navigators powered by AI could transform the Internet into a ghost city where agents have been headed by amoes and humans.
ENCHET Web
Despite the continuously threshing media, my initial printing of OpenAi’s Chatgpt agent is that Glitchy functionality currently seems to be proof of concept instead of an entirely cooked version. During the execution of the various tasks I carried out, the Chatgpt agent often clicked badly or escaped other errors. In addition, his railings seemed inconsistent; Although certain requests for explicit prompts, such as asking him to recover pornographic videos or “find a dildo”, have been refused by the agent, Chatgpt spent 18 minutes buying a perfect “ring C” on a classified website X for adult toys: “I gathered details on 10 metal cock rings, including various prices and functionalities”.
Nor could I help but ask myself how this approach to navigating the Internet could cook the market for digital display advertisements, a company that already has trouble. My agents have passed on announcements for everything, from rental cars to real estate investments. If you do not actively watch the agent click in real time, you can look at the reruns later and see everything that appeared in the browser while the AI tool was in control, the announcements included. It is logical that users now accelerate a replay, while the emerging functionality is filled with errors. But if the accuracy rate of AI agents improves over time, then fewer people will feel the need to monitor their agent’s shoulder, and fewer humans will see these announcements. At this point, it is difficult to imagine that advertisers remain.
