OpenAI has shared new details about dots, its always-on AI personal assistant. The product keeps working in the background, even when users are not giving it instructions.
The company released dots at its developer day in late September. Alexander Embiricos, who leads OpenAI’s enterprise product team, discussed the product at TechCrunch Disrupt this month.
dots is available in the ChatGPT desktop and mobile apps. It is open to subscribers on the Pro tier or higher, with plans starting at $100 a month.
Once users connect it to their calendar, email, computer and other tools, the agent tries to help them get work done. It can flag calendar conflicts and upcoming deadlines.
OpenAI is not alone in this market. The startup Instinct gained fast popularity this year, and Meta has heavily marketed its own assistant, Muse.
From Coding Tool to Personal Assistant
Embiricos said much of what dots does could already be done with Codex, OpenAI’s coding tool released early this year. However, Codex requires more work and hand-holding from the user.
He said the team focused on making the product feel intuitive. Framing the assistant as a single entity in one ongoing conversation was a key part of that effort.
The name is a play on the negative space inside OpenAI’s logo. Embiricos said he learned “very late” that Elon Musk owns the dot.com web address.
In TechCrunch testing, the assistant’s ability to think ahead was often useful but not always timely. Embiricos said that early in development, his own dot warned him about a calendar conflict just three minutes before it happened.
He expects the timing to improve. He also shared a story of an engineer at a drone delivery startup whose dot spotted an outage, analyzed it and suggested fixes before the engineer noticed.
Privacy and Payments
dots includes a feature called proactive research. On a regular schedule, it reviews what it knows about the user and looks for new ways to help.
OpenAI says dots has the same privacy controls as ChatGPT. Users can turn off model training and control what the agent can access on their computer.
Those controls are not very detailed, though. For example, the app will not accept read-only access to cloud drives.
Embiricos said OpenAI needs to build “super, super, super granular controls.” He added that the company received criticism at launch because the agent asked for permission often under its conservative default settings.
Hark, another startup building an AI assistant, says it does not collect data to support an advertising business. Meta’s Muse has faced criticism for the profiles it builds of users’ friends and family.
Embiricos also has a human assistant, because OpenAI’s corporate partners prefer dealing with a person. He said he never intends to meet with someone else’s agent.
Looking ahead, Embiricos said OpenAI is “deep in lots of conversations with payment processors.” Such partnerships could eventually let the assistant spend money on a user’s behalf.
