OpenAI Launches Dots Personal Assistant

OpenAI released dots, an always-on personal assistant, at its developer day in late September. Alexander Embiricos, who leads OpenAI’s enterprise product team, described its origins at TechCrunch Disrupt this month. The assistant is available in the ChatGPT desktop and mobile apps to Pro-tier subscribers and above; plans start at $100 a month. Once connected to a user’s calendar, email, computer and other tools, dots tries to help with work in the background, including when the user is not prompting it.

Embiricos said dots grew from a push to make AI assistants feel like a single entity and conversation. OpenAI had already spun ChatGPT Work out of agentic coding tools built for software engineers. Embiricos said much of dots’ functionality could be achieved with the Codex app, OpenAI’s coding tool released at the beginning of this year, but that doing so takes more work and guidance. He recalled celebrating when his wife began leaving her computer on overnight so her agents could continue working. He said making the product intuitive depends partly on how it is presented to users, rather than adding new technology.

The persistent agent is designed to work without repeated prompts. OpenAI gets more token usage from that activity, while users get an assistant that can look for ways to help. In TechCrunch’s testing, dots flagged calendar conflicts and upcoming deadlines, though not always soon enough. Embiricos recalled an early version warning him about a conflict only three minutes before it occurred. He expects timing to improve. He also described a drone-delivery startup engineer whose dot noticed and analyzed an outage before alerting the engineer with information to help resolve it.

Dots includes a feature OpenAI calls proactive research: on a schedule, it reviews what it knows from conversations and the resources a user has allowed it to access, then proposes ways to be more helpful. OpenAI says dots has the same privacy controls as ChatGPT, including the ability to turn off model training and control which computer resources it can access. TechCrunch reported that these controls are not granular and that the app can be pushy; for example, it refuses read-only access to cloud drives. Embiricos said OpenAI needs to build much more detailed controls. He also said the company initially set conservative defaults that led the assistant to ask for permission frequently, and that it could adjust them over time.

The article raises questions about whether personal AI agents can scale across businesses, websites and payment networks. Embiricos said OpenAI works closely with the wider ecosystem when it ships new ways for agents to interact with the world, and said the company is in discussions with payment processors. OpenAI’s corporate partners want to interact with a person, he said; Embiricos added that he has a human assistant and does not intend to meet another person’s agent. The article notes that payments access could be a significant ask for consumers, while its reporter said access to an email inbox already felt like a significant ask and a bank account was off limits for now.

OpenAI released dots, an always-on personal assistant, at its developer day in late September. Alexander Embiricos, who leads OpenAI’s enterprise product team, described its origins at TechCrunch Disrupt this month. The assistant is available in the ChatGPT desktop and mobile apps to Pro-tier subscribers and above; plans start at $100 a month. Once connected to a user’s calendar, email, computer and other tools, dots tries to help with work in the background, including when the user is not prompting it.

Embiricos said dots grew from a push to make AI assistants feel like a single entity and conversation. OpenAI had already spun ChatGPT Work out of agentic coding tools built for software engineers. Embiricos said much of dots’ functionality could be achieved with the Codex app, OpenAI’s coding tool released at the beginning of this year, but that doing so takes more work and guidance. He recalled celebrating when his wife began leaving her computer on overnight so her agents could continue working. He said making the product intuitive depends partly on how it is presented to users, rather than adding new technology.

The persistent agent is designed to work without repeated prompts. OpenAI gets more token usage from that activity, while users get an assistant that can look for ways to help. In TechCrunch’s testing, dots flagged calendar conflicts and upcoming deadlines, though not always soon enough. Embiricos recalled an early version warning him about a conflict only three minutes before it occurred. He expects timing to improve. He also described a drone-delivery startup engineer whose dot noticed and analyzed an outage before alerting the engineer with information to help resolve it.

Dots includes a feature OpenAI calls proactive research: on a schedule, it reviews what it knows from conversations and the resources a user has allowed it to access, then proposes ways to be more helpful. OpenAI says dots has the same privacy controls as ChatGPT, including the ability to turn off model training and control which computer resources it can access. TechCrunch reported that these controls are not granular and that the app can be pushy; for example, it refuses read-only access to cloud drives. Embiricos said OpenAI needs to build much more detailed controls. He also said the company initially set conservative defaults that led the assistant to ask for permission frequently, and that it could adjust them over time.

The article raises questions about whether personal AI agents can scale across businesses, websites and payment networks. Embiricos said OpenAI works closely with the wider ecosystem when it ships new ways for agents to interact with the world, and said the company is in discussions with payment processors. OpenAI’s corporate partners want to interact with a person, he said; Embiricos added that he has a human assistant and does not intend to meet another person’s agent. The article notes that payments access could be a significant ask for consumers, while its reporter said access to an email inbox already felt like a significant ask and a bank account was off limits for now.