OpenAI Turns Its Coding Tools Into ‘Dots,’ An Always-On Assistant That Runs In The Background

OpenAI this fall introduced Dots, an always-active assistant built into ChatGPT’s apps that stays connected to users’ calendars, email and devices to help manage everyday work. The feature, announced at the company’s developer event and discussed this month by Alexander Embiricos, who heads OpenAI’s enterprise product group, is available to subscribers at the Pro level and above, with plans beginning at $100 a month. The launch puts OpenAI alongside startups and big tech firms that have pushed animated or branded agents as the consumer face of large AI models.

Dots grew out of tools OpenAI originally built for programmers. Embiricos described to a TechCrunch event audience how the company repackaged backend capabilities into a persistent agent aimed at nondevelopers, moving from a command-driven coding assistant to something that keeps working without constant prompting. The name is a nod to the firm’s logo, he said, and was chosen with the idea of creating a familiar, widely used interface rather than a niche product. He also mentioned discovering late in the process that a rival entrepreneur had registered a related domain, a small hiccup in a larger rollout.

The defining trait of Dots is proactivity: once granted access, the agent will scan schedules, inboxes and other linked tools and surface suggestions or alerts. That approach drives more model usage behind the scenes and, in theory, reduces the manual steps a user must take. Early experiments show value but not perfection — Embiricos acknowledged instances where alerts arrived only moments before a meeting, and said the timing and judgement of those nudges will be refined. He also shared an account of an engineer at a delivery drone company whose agent detected and diagnosed an outage before the human on call had noticed, then delivered a concise path toward remediation.

That usefulness depends on how much information the assistant can access, which raises familiar privacy questions. OpenAI has built a capability it calls proactive research: at intervals the agent reviews what it knows about you, evaluates the data it can query, and suggests ways to be more helpful. Users can opt out of having their exchanges used to train OpenAI’s models and can limit which apps the assistant may reach, but current controls lack fine-grained options and the app currently resists read-only access for some cloud storage setups. Embiricos acknowledged those shortcomings and said the company started with conservative defaults that asked for permissions frequently, with plans to tune settings and add more detailed controls.

Beyond individual privacy choices, Dots highlights a broader trade-off facing the industry: convenience versus exposure. Competitors have drawn criticism for the profiles their systems assemble on users’ social networks and contacts, and businesses aren’t uniformly prepared to accept autonomous agents acting on customers’ behalf. Embiricos said many corporate partners still prefer human contact, and he himself keeps a human assistant for meetings, noting he doesn’t expect to meet with another firm’s agent in place of a person.

There’s also a technical and regulatory hurdle to wider adoption. For agents to act on behalf of users — scheduling, ordering, or even making payments — they must be integrated with services that weren’t built for machine-initiated traffic. OpenAI confirmed it is in talks with payment processors and other ecosystem players, seeking close collaboration so agents operate safely and reliably. Industry observers say those conversations will determine whether assistants can move beyond simple alerts to perform higher-stakes tasks.

Privacy lawyers and enterprise IT leaders contacted for this article urged caution and standards. One data-privacy researcher warned that centralized, always-on assistants could concentrate sensitive signals unless governance and audit logs are robust. An enterprise technology director said the productivity gains are real — a well-configured agent can shave hours from routine workflows — but firms will demand contractual guarantees and technical safeguards before broad deployment. For now, Dots represents an early but notable step toward assistants that do more of the thinking for users, leaving the tougher questions about control, oversight and payment access to be worked out.