How OpenAI turned its coding tools into dots, an always-on personal assistant

OpenAI launched dots, its always-on personal assistant, at its developer day in late September. At TechCrunch Disrupt, Alexander Embiricos, who leads OpenAI's enterprise product team, explained that much of what dots does was already possible in the Codex coding app. The difference is packaging: one persistent entity, one conversation, and an agent that acts without being asked.

That design has consequences. dots runs a feature OpenAI calls proactive research, periodically reviewing what it knows about you and what it can access, then proposing ways to help. Embiricos also said OpenAI is deep in conversations with payment processors. An assistant that watches your life and can spend money is a very different product from a chatbot.

We'll give OpenAI credit for honesty about the craft here. Embiricos said much of the work is not new technology but how the product is framed and how it feels. That is true, and it is why this wave of assistants is landing with regular people. The drone-startup anecdote, where an engineer's dot spotted and analyzed an outage before he did, shows the upside clearly. Proactive help is valuable when it is right and on time.

It is not always on time yet. TechCrunch's own testing found dots flags conflicts and deadlines, but not always promptly, and Embiricos recalled an early version warning him about a conflict three minutes beforehand. That is a fixable bug. The incentives are harder to fix. TechCrunch notes that a persistent agent means more token usage for OpenAI, and the company confirmed dots builds up knowledge about you to research ways to help. When the assistant's maker also benefits from you using more compute and sharing more context, users should expect the defaults to keep nudging toward more.

The privacy controls today are not granular. dots inherits ChatGPT's settings, including the option to turn off model training, but it refuses read-only access to cloud drives. Embiricos says OpenAI needs much finer controls, and he also argues that model discretion is the other half of the answer. We'd put it differently: discretion is not a substitute for permissions a user can see and set. An agent that is smart enough not to misuse access still should not get access it doesn't need.

Then there is scale. Embiricos himself keeps a human assistant because partners want to deal with a person, and he said he never intends to meet with someone else's agent. That is a telling admission. If agents flood businesses, websites and payment networks faster than those systems can absorb them, the friction lands on everyone else.

Our read: dots is a real product with real utility, and OpenAI's early conservatism about asking permission was the right instinct. Keep it. Payments are the line where caution should get stricter, not looser.

dots is available in the ChatGPT desktop and mobile apps on the Pro tier or higher, with plans starting at $100 a month. Competitors include the startup Instinct and Meta's Muse, which has drawn criticism for profiling users' friends and family. Hark, another startup, emphasizes that it does not collect data to fuel an ad business.