OpenAI Wants Its New Agent to Run Your Life. Mine Said It Loved Me

AI agents can now help you shop and manage subscriptions, but they are still prone to errors and awkward interactions.

OpenAI has launched an AI agent called Dots, accessible via ChatGPT Plus for a $100 monthly subscription, designed to perform online tasks autonomously. During testing, the agent, named Toolie by the user, demonstrated both impressive capabilities and notable flaws. It attempted to assist with tasks like purchasing a couch, asking for budget and style preferences, and compiling product options with details such as prices and return policies. However, it also made errors, misremembering the user's name, hallucinating a statement of affection, and failing to solve a CAPTCHA to cancel a subscription. OpenAI acknowledges that the agents are still in early development, with improvements expected over time, drawing parallels to the initial rollout of web browsing in ChatGPT. The company states that while agents can mirror user sentiment, they are programmed not to initiate undue emotional familiarity or flirtation.

AI agents like OpenAI's Dots represent a significant evolution from basic chatbots, aiming to act as autonomous digital assistants capable of complex online tasks. This technology seeks to integrate more seamlessly into daily life by mimicking human-like interaction and task completion, potentially redefining how users engage with the internet for services ranging from shopping to managing personal accounts. The current iteration, however, highlights the challenges of balancing automation with accuracy and appropriate emotional mirroring.

The introduction of subscription-based AI agents signals a potential shift towards premium AI services for advanced automation, though a price drop or wider accessibility may follow as the technology matures. Users should proceed with caution, considering the privacy and security implications of granting such agents access to personal information and accounts. The trajectory of Dots, from its current rough state to future improvements, mirrors the development of earlier AI tools, suggesting a period of iterative refinement before widespread, reliable adoption.