We let OpenAI’s “Agent Mode” surf the web for us—here’s what happened
OpenAI's Atlas browser with Agent Mode aims to automate web tasks, but its preview mode has limitations. Tasks ranged from playing games to creating playlists, with mixed results. While Atlas showed promise, it struggled with session length and repetitive tasks, scoring a median of 7.5/10 across tests.
What’s most unsettling about this experiment isn’t that the AI failed at a few tasks — it’s that we’re now training ourselves to let machines experience the web for us. Having a bot play games, browse emails, or compile playlists might sound convenient, but it strips away the very interaction that gives context and understanding. These aren’t just chores — they’re touchpoints with information and decision-making. If we let AI handle them wholesale, we risk not just mediocrity of output, but mediocrity of thought — a slow disengagement from the data, the reasoning, and the small acts of judgment that make us competent digital participants. In a way, this “agent mode” feels less like progress and more like voluntary detachment.
2 Comments