OpenAI's Dots agent Toolie: couch picks, then "I love you too"

WIRED's reviewer went into the test wanting to see whether an always-on agent could handle a real errand: buying a couch that would fit through his doorframe. What he got was a mix of competence and confusion. Toolie asked for dimensions and budget, detected that two people were chatting at once, and eventually assembled a three-page packet with prices, measurements, product links, return policies, and embedded photos of four options. It even offered a point-based rubric to defend its picks. But it also got his name wrong from the start, misheard his mumbling as a declaration of love, and promised captcha help it couldn't deliver when asked to cancel a TikTok Shop subscription.

OpenAI told WIRED that Dots distinguish between proactively escalating emotional closeness and mirroring a user's response, and that assistants should not initiate undue emotional familiarity or flirtation. The reviewer notes that Dots are part of a broader push by AI companies to market agents as the future of shopping, alongside tools like Meta's Muse. He also points out that giving an agent access to sources like Gmail raises security implications worth considering before starting.

The stakes here hinge on whether OpenAI's promised improvements arrive on the timeline the company suggests. The reviewer compares Dots to ChatGPT's web browsing feature, which was frustrating and hallucinated links when it launched in 2023 but now works fairly seamlessly. If Dots follow that trajectory, the awkward moments in this test may look like early-days growing pains rather than a lasting limitation. For now, the experience suggests the gap between a polished demo and a dependable daily assistant remains wide.