OpenAI shipped dots at its developer day in late September. The interesting part is not that the assistant exists but how it is built — and what access it demands.
The facts first. dots lives in the ChatGPT desktop and mobile apps for subscribers on the Pro tier or higher, plans from $100 a month. Once connected to your calendar, email, computer and other tools, the agent tries to get your work done. Alexander Embiricos, who leads OpenAI's enterprise product, concedes that much of the functionality was already reachable through the Codex coding app — with more hand-holding. dots is, to a real degree, a reframing of coding agents for people who do not code.
The mechanism is called proactive research. On a cadence, dots thinks about what it knows about you, what came up in the conversation, and what you have given it access to — then researches what could be more helpful and proposes it unprompted. For OpenAI that means more tokens burned; for the user an assistant that is, in Embiricos's phrase, "sort of thinking about you." In TechCrunch's testing the forward thinking was apparent and often useful, but not always timely: Embiricos recalls his own dot flagging a calendar conflict three minutes before the meeting. "I only have three minutes!"
The most revealing detail, though, is about permissions, not features. dots inherits ChatGPT's privacy controls: training can be turned off, resource access adjusted. But the controls are not granular, and the app is pushy — it refuses to accept read-only access to cloud drives. If you want dots working over your files, it gets write access. Embiricos says the company needs to build "super, super, super granular controls." He also admits the launch defaulted to very conservative behaviour and asked permission constantly, drawing criticism they intend to tune away.
The ecosystem question gets a candid answer. Embiricos says he never intends to attend a meeting with someone else's agent, and he keeps a human assistant because OpenAI's corporate partners want to deal with a person. On money he is more forward-looking: OpenAI is "deep in lots of conversations with payment processors." An assistant that buys things on your behalf is no longer a product idea; it is a negotiation.
Even the name is strategy. dots plays on the negative space inside OpenAI's logo and was chosen, per Embiricos, to feel "like just something that could eventually be ubiquitous." He learned that Elon Musk owns the dot.com domain, he says, "very late."
Why it matters: the always-on assistant now has three serious entrants — the startup Instinct, Meta's Muse, and OpenAI — and dots shows where the real constraint sits. Not model capability, but the permission model. An assistant that will not work over your files without write access is asking for more authority than the task requires. That single refusal is the most concrete engineering detail in the interview.
