We let OpenAI’s Agent Mode navigate the web for us

On Tuesday, OpenAI announced Atlas, a new web browser with ChatGPT integration, which allows users to “chat with a page,” as the company describes it. But Atlas also goes beyond the usual LLM exchange with Agent Mode, a “preview mode” feature that, according to the company, can “do the work for you” by clicking, scrolling, and reading multiple tabs.

The “Agentic” AI is nothing new, of course; OpenAI released a preview of the Operator web browsing agent in January and unveiled the more general-purpose “ChatGPT” agent in July. Even so, highlighting this capability in a major product launch like this, even in preview mode, indicates a clear push to make this type of system available to end users.

I wanted to test Atlas’s Agent Mode to see if it could actually save me time on the tedious online tasks I perform daily. In each case, I’ll describe a web problem, present the Agent Mode instruction I devised to try and solve it, and describe the results. My final evaluation will rate each task on a 10-point scale, where 10 is “did exactly what I wanted without any problems” and 1 is “a complete failure.”