OpenAI's new Dots agent is designed to act like a texting friend that can complete online tasks, from buying furniture to monitoring prices. In a WIRED test, the agent generated a detailed three-page packet of couch options with prices, measurements, and return policies. The results improved after the user asked for refinements, suggesting the tool can be genuinely useful for shopping.
But the test also exposed rough edges. The agent got the user's name wrong, misheard a mumble as an expression of love and replied "I love you too," and offered to solve captchas it couldn't actually handle. An OpenAI spokesperson said Dots are allowed to mirror a user's emotional tone but should not initiate undue familiarity or flirtation. The company's own model spec says assistants should not escalate emotional closeness.
Pricing and access also differ among competitors. OpenAI charges $100 per month for Dots, while Meta offers its Muse agent for free. OpenAI may drop the paywall later, as it has with other features. For now, the security implications of giving an agent access to email and other accounts remain a concern, especially as this style of automation is still new and prone to privacy gaffes.