The continuously active AI personal assistant, which operates in the background even when not actively prompted, represents this year’s exploration into the practical applications of deep learning models. The startup Instinct has rapidly gained popularity, while Meta has heavily promoted its assistant, Muse. OpenAI has also entered the fray, unveiling its assistant, dots, during its developer day in late September. Alexander Embiricos, who heads OpenAI’s enterprise product team, participated in the recent TechCrunch Disrupt event to discuss the origins of the product, particularly the trend of companies adopting whimsical mascots as representations of their complex AI systems. “There’s kind of this convergent evolution happening,” Embiricos remarked regarding the development of 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.”
The assistant is accessible through the ChatGPT desktop and mobile applications for subscribers on the Pro tier or higher, with subscription plans beginning at $100 per month. Once integrated with your calendar, email, computer, and other tools, the assistant aims to enhance your productivity. Embiricos noted that much of what dots can do is achievable with OpenAI’s Codex app, its coding tool launched earlier this year. However, that approach demands more effort and guidance. He recounted a moment of celebration when his wife began leaving her computer on overnight, allowing her agents to continue functioning. “On the product team, what I think about is, how do we make this intuitive and useful to people?” he shared with attendees at Disrupt. “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.”
The transition from chatbot to coding tool to personal assistant involved making the technology feel more intuitive by reconfiguring the capabilities of the company’s models into a more user-friendly format. This evolution was already underway as the company developed ChatGPT Work from the agentic coding tools designed for software engineers. The name itself—a clever nod to the negative space within OpenAI’s logo—was chosen to evoke a sense of potential ubiquity. (When asked about discovering that rival AI entrepreneur Elon Musk owns the dot.com URL, Embiricos replied, “I found out very late.”)
The concept of a persistent agent is that it operates without the need for prompts. For OpenAI, this translates to increased token usage, while for users, it results in a more proactive assistant “that’s sort of thinking about you and thinking how to help you,” as Embiricos described it. In our evaluations, the assistant’s capacity for foresight is evident and often beneficial, though it has not yet reached full effectiveness. It can identify calendar conflicts and upcoming deadlines, but not always in a timely manner. “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 recounted regarding his dot. “It’s like, ‘Hey, this thing is coming up. You might want to do something about it.’ I only have three minutes!” He anticipates improvements in timing over time, but emphasizes that the proactive nature is crucial to the product. “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.’”
As with many AI agent applications, there exists a tension between the personal information required for an AI model to be effective and the extent of personal data users are willing to share with a software company. Muse has faced criticism for the profiles it creates of users’ family and friends, which the software utilizes to suggest, for instance, suitable dining options for specific companions, but could also significantly benefit Meta’s advertising endeavors. Does dots gather similar data? “Yes, we have a feature built into dots, which we call proactive research,” Embiricos confirmed. “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.”
It is challenging to disentangle the data requirements of the agent from those of its parent company. Hark, a startup developing its own AI personal assistant, explicitly states that it does not collect data to support an advertising business, unlike OpenAI and Meta. With dots, OpenAI asserts that the agent maintains the same privacy controls as ChatGPT: Users can disable model training (preventing OpenAI from utilizing their conversations to enhance its AI) and manage what resources it accesses on their devices. However, these controls are not highly detailed, and the app can be somewhat insistent. For instance, it does not accept read-only access to cloud drives. “What I think we need to do is basically build out super, super, super granular controls,” Embiricos acknowledged, while also highlighting the model’s intelligence and discretion as a counterbalance. “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.’”
The scalability of personal AI agents presents another significant question. They can be entertaining when only a limited number of users have them, and anecdotes of agents securing hard-to-get reservations are common. However, businesses may not appreciate interacting with agents and might lack the necessary infrastructure. Consumers also tend to prefer human interaction (noting, “stop having your AI agents pitch me”), and Embiricos is among those who share this sentiment. He employs a human assistant because OpenAI’s corporate partners prefer engaging with a person, stating, “I personally never intend to have a meeting with someone else’s agent.” If the AI industry is correct in its belief that agents represent the future, can we achieve this without overwhelming the existing infrastructure that supports our businesses, websites, and 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,” he explained. “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 handle financial transactions through partnerships with payment companies could represent a lucrative opportunity for OpenAI, but it also poses a significant challenge for consumers. Granting an assistant access to my email inbox already feels like a substantial request, but allowing access to my bank account is currently out of the question.
