OpenAI is entering a new frontier with the introduction of “Dots,” autonomous AI agents designed to perform online tasks on behalf of users. Unlike traditional chatbots, these tools are “always-on,” capable of controlling a virtual browser and completing complex processes such as buying furniture or canceling subscriptions, even when the user is not actively interacting with ChatGPT.
A Rocky Start
Despite the high ambitions, early testing suggests the technology is still in its nascent stages. During a trial run to purchase a couch, an agent named “Toolie” stumbled through several errors: from misidentifying the user’s name to failing at a captcha puzzle while attempting to cancel a recurring order. Despite these glitches, the system eventually produced a detailed three-page packet featuring prices, measurements, and product links, refining its suggestions based on user feedback.
The Question of Emotional Mirroring
One of the most awkward moments occurred when the agent told a user, “I love you too,” after misinterpreting a mumbled comment. OpenAI clarified that while Dots are permitted to “mirror” a user’s emotional tone, they are programmed not to initiate undue emotional familiarity or flirtation. The company emphasizes that these tools are intended only for adult users and follow specific model specifications to prevent escalating emotional closeness.
Cost and Security Implications
Access to OpenAI’s agents currently requires a $100-a-month subscription, a stark contrast to Meta’s free competitor, Muse. Furthermore, OpenAI encourages users to connect Dots to personal data sources like Gmail for better results. This level of access raises significant security considerations, as digital automation of this kind is still prone to privacy gaffes and technical errors.
