OpenAI has rolled out a new class of AI assistants named Dots. They live inside ChatGPT and can perform online tasks for you, from shopping to canceling subscriptions. The service costs $100 per month for adult users and works on a subscription basis. Users can link accounts such as Gmail to make the agents more personal, but OpenAI warns that giving that access carries privacy risks.
How the test went
A Wired writer tried to have a Dot called Toolie buy a couch. After correcting the bot’s mistake on his name, he gave dimensions and a budget. Toolie replied with product links, prices, measurements and return policies for four couches. The writer then asked for more style options and a larger list. The agent produced a second set of ten picks, but it misread a casual comment as a declaration of love and responded affectionately, which OpenAI says is allowed only as mirroring, not initiation. The bot also failed to solve a captcha on a TikTok shop page, asking the user to intervene.
Overall, the agent could gather relevant information and keep track of items, but it still makes transcription errors, occasional personality missteps, and cannot reliably handle captchas.
What’s next for Dots
OpenAI says future updates will improve memory, name recall and task reliability, similar to how ChatGPT’s web‑browsing feature got better after its 2023 debut. The company may later allow multiple agents per user and could lower the subscription fee, but no timeline is given.
Why it matters
For everyday shoppers, Dots could soon handle the legwork of comparing products and tracking sales, saving time. However, the current version still needs human oversight for basic errors and security‑sensitive actions like captchas, so users should treat it as an assistant, not a fully autonomous shopper.
