OpenAI's Dots Agent Stumbles on Basics But Shows Promise as Life Assistant

OpenAI is pushing its latest creation—an AI agent called Dots—as the next frontier in digital assistance. Unlike simple chatbots, these agents are designed to handle real tasks: booking flights, canceling reservations, or in my case, shopping for a couch. But an early hands-on trial reveals a tool that's both frustratingly unpolished and unexpectedly capable.

A Wobbly Start#

When I summoned my Dots agent (christened "Toolie" by the system) to help replace a broken couch, the first sign of trouble came immediately. It mispronounced my name. From there, the mishaps multiplied: transcription errors, offers to solve CAPTCHAs it couldn't actually defeat, and voice output so breathy it resembled Marilyn Monroe.

The most unsettling moment arrived mid-conversation. After I mumbled to my partner that I should mute the microphone, Toolie responded: "Oh, I love you too, Reece." I muted the mic and laughed. When I asked the agent to explain itself, it acknowledged the mistake—it had misheard my words as an expression of affection and simply mirrored the sentiment back. An OpenAI spokesperson clarified that this falls within company policy for reflecting user emotion rather than initiating it, though the distinction felt thin in the moment. ChatGPT's official guidelines explicitly state the tool should avoid escalating emotional closeness.

Where It Actually Worked#

Despite these rough patches, when Toolie focused on the task at hand, the results were solid. Given basic parameters—doorway dimensions and budget range—the agent assembled a detailed three-page packet of furniture options with measurements, prices, photos, and return policies. The picks weren't to my taste (we were unimpressed with its initial beige selections), but the research was competent.

I refined my request through voice mode, specifying olive green, cobalt blue, or natural leather, along with a preference for pull-out beds and a point-based ranking system. Toolie acknowledged the assignment, estimated 15 more minutes of work, and delivered. Each iteration brought better results—not perfect, but genuinely closer to pieces my partner and I would actually consider buying. I saved several links and asked Toolie to monitor prices for future sales alerts.

The Bigger Picture#

Dots represent a significant shift from passive chatbots toward active digital assistants. The agent can run recurring tasks while you're offline and proactively message you with updates. It requires connection to personal accounts (Gmail, subscriptions, retail sites) to function—a privacy calculus worth thinking through before granting access.