OpenAI Turns Coding Tools Into Personal Assistants With Dots

The always-on AI personal assistant has arrived as this year’s answer to how people will actually use deep learning models. Following viral interest in startups and Meta’s Muse, OpenAI launched its own entry named dots at its developer day in late September. Alexander Embiricos, who leads OpenAI’s enterprise product team, explained at TechCrunch Disrupt that the product evolved from the company's agentic coding tools.

Instead of requiring the manual hand-holding of earlier software, dots functions as a single, continuous conversation designed to feel intuitive. Available to subscribers paying one hundred dollars a month or more, the desktop and mobile assistant connects to email, calendars, and computers to anticipate user needs. While testing shows the assistant can think ahead, its timing remains inconsistent, occasionally flagging schedule conflicts only minutes before they occur.

Under the hood, a proactive research feature periodically reviews stored data and connected resources to generate helpful suggestions. Yet this persistent monitoring highlights a familiar tension regarding how much personal data users must expose to software providers, even with privacy controls that allow turning off model training. Our read is that OpenAI is betting user tolerance for data exposure will expand if the utility of a background agent outpaces the friction of its guardrails.

While company defaults remain conservative and ask for frequent permissions, the push toward deeper integration is already testing boundaries.

Embiricos noted that developers are in talks with payment processors to let agents spend money, crossing a line from managing schedules to executing financial transactions.

Whether businesses and users want to interact with autonomous software remains an open question.

Embiricos himself admitted he never intends to hold a meeting with another person’s agent, pointing to a broader hesitation across the ecosystem.

As agents transition from novelty tools to persistent background entities, the real test will be whether infrastructure and human patience can scale alongside them.