The always-on AI personal assistant—software that quietly keeps working in the background even when users are not actively prompting it—has rapidly emerged as the tech industry’s defining answer to how people will practically integrate deep learning models into their daily lives. Over the past year, the artificial intelligence landscape has seen massive shifts in this direction. A startup named Instinct became an overnight sensation with its background agent software, and social media giant Meta threw the full weight of its marketing apparatus behind its own assistant, Muse.
OpenAI was determined not to be left behind in this race. At its developer day in late September, the company officially released "dots," its highly anticipated personal assistant offering. Alexander Embiricos, who leads OpenAI’s enterprise product team, joined journalists at TechCrunch Disrupt this month to unpack the origins of the product. The conversation naturally began with a fundamental question currently facing the entire tech sector: why are so many major technology companies suddenly turning to cutesy mascots and minimalist avatars to represent their trillion-parameter matrix multiplication machines?
"There’s kind of this convergent evolution happening," Embiricos said, describing the industry-wide push to build sophisticated AI assistants. "As we were working on the product, we realized if we could frame this as, it’s an entity and it’s a single conversation, not many different conversations… it starts to become intuitive."
Currently, the assistant is available within the ChatGPT desktop and mobile applications exclusively for subscribers on the Pro tier or higher, with subscription plans starting at $100 a month. Once the software is securely connected to a user’s calendar, email client, computer files, and other digital tools, the agent functions as a proactive worker designed to help streamline daily workloads.
Embiricos pointed out during the panel that users could technically achieve much of the functionality found in dots by utilizing OpenAI’s Codex app, a dedicated coding tool the company released earlier this year. However, leveraging Codex for broad personal assistance requires considerably more manual effort and hand-holding from the user. He recalled celebrating a minor milestone at home when his wife began leaving her computer running all night so that her AI agents could continue working while she slept.
"On the product team, what I think about is, how do we make this intuitive and useful to people?" he told Disrupt attendees. "A lot of that isn’t building some new piece of tech into the product; it’s how you frame it to the user and how it feels."
Making the technology feel intuitive ultimately meant repackaging the raw functionality of the company’s powerful underlying models into something much more accessible to everyday consumers. This marked the continuation of a strategic path OpenAI was already traversing, having previously spun ChatGPT Work out of the agentic coding tools originally built for software engineers. Even the product’s name—dots, which serves as a clever play on the negative space nestled inside OpenAI’s corporate logo—was chosen deliberately. According to Embiricos, it is intended "to feel like just something that could eventually be ubiquitous." (Asked by an audience member when he discovered that rival AI entrepreneur Elon Musk conveniently owns the dot.com URL, Embiricos offered a candid response: "I found out very late.")
The core philosophy behind a persistent agent is that it operates autonomously without requiring continuous prompting from the user. For OpenAI, this shift translates into significantly higher token usage across its infrastructure. For users, however, it promises a much more proactive assistant—one that, in Embiricos’s words, is "sort of thinking about you and thinking how to help you."
In practical testing of the software, the assistant’s capacity to think ahead is frequently apparent and often genuinely useful, though it is clear the technology is still maturing. The agent is capable of flagging calendar conflicts and upcoming deadlines, for instance, but it does not always deliver these warnings in the most timely fashion.
"The first time it told me about one of those [conflicts]—this is like early in development—it told me three minutes before the thing," Embiricos said of his own experience with his dot. "It’s like, ‘Hey, this thing is coming up. You might want to do something about it.’ I only have three minutes!"
Embiricos expects the timing and predictive accuracy of the software to improve steadily over time, but he maintains that the proactive approach remains the cornerstone of the product’s value proposition. He shared an anecdote that highlighted the potential of this design philosophy. "I was hearing about an engineer at a drone delivery startup, and they were saying, ‘We had this outage, and by the time I noticed that the outage was going on, my dot had already noticed the outage was happening, had already analyzed the outage, and had come to me with useful information for how we could resolve the outage.’"
Giving Dot the Keys
As is the case with virtually all consumer-facing AI agent use cases, the development of dots brings to light a persistent tension. There is an inherent friction between the vast amount of personal information an AI model needs to possess to genuinely be useful, and the exact degree of personal data a user is willing to expose to a large software corporation. Meta’s assistant, Muse, has previously faced significant public criticism for the detailed psychological and social profiles it constructs of a user’s friends and family. While this software uses the data to suggest appropriate restaurants for a specific pal, it simultaneously builds an incredible commercial asset that feeds Meta’s lucrative advertising ecosystem. Naturally, this raises questions regarding what kind of data dots collects and utilizes.
"Yes, we have a thing that we built into dots, which we call proactive research," Embiricos explained. "On a cadence, it will think about what it knows about you, what is the conversation that it’s had, and then what have you given it access to that it can think about, and then it will go and research, ‘What could I do to be more helpful to you?’ And then, basically, it will propose that."
Drawing a clear boundary between the data requirements of an independent agent and the corporate interests of its parent company remains a complex challenge for consumers. Other industry players are positioning themselves differently; for example, Hark, a startup currently developing its own competing AI personal assistant, explicitly emphasizes that it does not harvest user data to fuel an advertising business in the manner practiced by corporate giants like OpenAI and Meta.
Regarding dots, OpenAI maintains that the agent operates under the exact same privacy controls established for standard ChatGPT interactions. Users have the option to turn off model training—effectively stopping OpenAI from utilizing their conversations and inputs to further train its foundational AI models—and retain control over which resources on their computer the agent can access, and to what extent.
Nevertheless, these privacy controls are not yet granular enough for every user’s comfort, and the application itself can occasionally feel overly assertive. For instance, the assistant currently refuses to accept read-only access to cloud drives, demanding broader permissions to function effectively.
"What I think we need to do is basically build out super, super, super granular controls," Embiricos acknowledged, though he also pointed to the model’s advanced intelligence and discretion as the necessary balancing factor on the other side of the coin.
"This is actually something we kind of got some flack about when we launched," Embiricos noted. "We had really conservative defaults over what the agent would do, and it would ask for permission a lot. I remember knowing that we were going to see that feedback, but thinking, ‘This is okay, we can tune that over time.’"
Can Personal AI Agents Scale?
Beyond data privacy concerns, the other major question surrounding the widespread adoption of AI agents is whether they can successfully scale across society. Persistent agents can seem like a novelty when only a select few people possess them, and anecdotes of clever bots successfully securing hard-to-get restaurant reservations frequently make the rounds online. However, traditional businesses do not necessarily enjoy interacting with automated agents and are frequently ill-equipped to handle them at scale. Furthermore, everyday people often dislike dealing with them as well—a sentiment echoed by Embiricos, who remarked that he frequently has to tell people to stop having their AI agents pitch him.
In fact, Embiricos relies on a human assistant himself because OpenAI’s corporate partners prefer interacting with a person when scheduling and coordinating. "I personally never intend to have a meeting with someone else’s agent," he admitted.
If the broader artificial intelligence industry proves correct in predicting that autonomous agents represent the undeniable future of computing, the critical question remains whether society can reach that future without completely overwhelming the existing digital and physical infrastructure that modern civilization runs on—ranging from brick-and-mortar businesses and informational websites to complex global payment networks.
"At OpenAI… we’re not always moving the fastest in terms of shipping the next ability for the agent to work with the world," Embiricos said. "When we do, we’re doing it really closely in collaboration with the rest of the ecosystem. I probably can’t say too much here, but we’re deep in lots of conversations with payment processors."
Allowing an AI assistant to autonomously spend money through official partnerships with major payment companies would represent a potentially lucrative industry first for OpenAI, but it also asks a tremendous leap of faith from ordinary consumers. For many users, granting an assistant read-and-write access to a personal email inbox already feels like an immense boundary to cross, keeping bank accounts and financial portfolios strictly off-limits—at least for now.
