The always-on AI assistant is emerging as one of this year’s biggest ideas in AI. Instead of waiting for users to type a prompt, these assistants can work quietly in the background and step in when needed. Instinct quickly grabbed attention, while Meta threw its marketing weight behind Muse, signaling a growing push to make AI a more natural part of everyday life.

OpenAI wasn’t about to sit out the race for AI assistants. The company introduced dots, its own take on the concept, at its developer event in late September. At TechCrunch Disrupt this month, Alexander Embiricos, who leads OpenAI’s enterprise product team, shared how the idea came together. He also addressed a curious trend in the AI industry: why so many companies are giving powerful AI systems cute mascots and friendly personalities.

“There’s kind of this convergent evolution happening,” Embiricos said, describing the growing interest in AI assistants. As the team developed the product, they realized that treating the assistant as a single entity you could have one ongoing conversation with, rather than juggling multiple separate chats, made the experience feel more natural and intuitive.

The assistant is available through ChatGPT’s desktop and mobile apps for Pro subscribers and above, with plans starting at $100 per month. Once connected to tools such as your calendar, email, and computer, it can work in the background to help manage tasks and get things done.

According to Embiricos, much of what dots does is already possible with OpenAI’s Codex app, the company’s coding tool released earlier this year. The difference is that Codex often requires more setup and user guidance. He recalled being excited when his wife began leaving her computer running overnight so her AI agents could continue working while she slept.

For Embiricos, that experience highlights the bigger challenge: making AI assistants useful without requiring users to figure out all the technical details themselves.

“What I think about on the product team is, how do we make this intuitive and useful to people?” he told the Disrupt audience. “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.”

From Chatbot to Coding Tool to Personal Assistant

Making AI feel more natural to use isn’t just about improving the technology. It’s also about presenting existing capabilities in a way that people can easily understand and use in their everyday lives.

OpenAI has been moving in this direction by building on the AI tools it originally developed for software engineers and turning them into something more accessible. ChatGPT Work grew out of that effort, bringing capabilities from its agentic coding tools to a broader range of tasks.

Even the name, dots, reflects the idea. Inspired by the empty spaces in OpenAI’s logo, it was chosen to suggest something simple that could eventually become a familiar part of everyday life. As Embiricos explained, the goal was for it “to feel like just something that could eventually be ubiquitous.”

There was one small surprise along the way. When asked when he discovered that rival AI entrepreneur Elon Musk owned the dot.com URL, Embiricos admitted, “I found out very late.”

The Promise and Challenges of an Always-On AI Assistant

The idea behind a persistent AI agent is simple: it shouldn’t have to wait for you to give it instructions every time. Instead, it should keep an eye on things in the background and step in when it thinks it can help. For OpenAI, that also means more token usage, while users get a more proactive assistant that, as Embiricos described it, is “sort of thinking about you and thinking how to help you.”

In our testing, dots showed that it could be useful to have an assistant that thinks ahead, although the experience still needs some refinement. It can flag calendar conflicts and upcoming deadlines, for example, but sometimes the alerts arrive too late to be of much help.

“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 recalled. “It’s like, ‘Hey, this thing is coming up. You might want to do something about it.’ I only have three minutes!”

He expects the timing to improve as the product develops, but sees this kind of proactive help as central to its appeal.

To illustrate what that could look like, Embiricos shared a story about an engineer at a drone delivery startup. The engineer reportedly said that when the company experienced an outage, their dot had already detected the problem, analyzed what was happening, and suggested ways to resolve it before the engineer even noticed the issue.

Giving Dot Access to Your Digital Life

For an AI assistant to be genuinely useful, it needs context. Access to your emails, calendar, files, and other personal information can help it make better suggestions. But that raises an important question: how much access should we give an AI company to our digital lives?

Meta’s Muse has faced criticism over the personal profiles it builds around users’ family members and friends. That information could help an assistant recommend a restaurant for a particular outing, but it also raises concerns about how personal data might benefit Meta’s advertising business. So, does OpenAI’s dots work in a similar way?

Embiricos explained that dots includes a feature called proactive research. At regular intervals, the assistant considers what it knows about a user, their previous conversations, and the resources they have allowed it to access. It then looks for ways to be more helpful and suggests possible actions.

The challenge is separating the data an AI assistant needs to perform well from the commercial interests of the company operating it. Hark, another startup developing a personal AI assistant, has highlighted its distinction from companies such as OpenAI and Meta by saying it does not collect data to support an advertising business.

OpenAI says dots follows the same privacy controls as ChatGPT. Users can disable model training, which prevents their conversations from being used to improve OpenAI’s models, and manage the resources the assistant can access on their computers.

However, those controls have limitations. The app can be demanding about permissions and, for example, refuses to accept read-only access to cloud drives.

Embiricos acknowledged that better controls are needed. “What I think we need to do is basically build out super, super, super granular controls,” he said.

He also pointed to a balancing act between giving users control and making the assistant capable enough to be useful. When dots first launched, its default settings were deliberately conservative, and it frequently asked users for permission before acting. Embiricos said the team expected some criticism and believed those settings could be adjusted over time.

Can Personal AI Agents Scale?

Beyond privacy, there’s another challenge for AI assistants: getting the rest of the world to work with them.

AI agents might make it easier to find restaurant reservations or handle routine tasks, but businesses aren’t necessarily equipped to interact with automated assistants. People may also be uncomfortable receiving messages from bots acting on someone else’s behalf. Embiricos himself isn’t a fan of that approach.

He also has a human assistant, partly because OpenAI’s corporate partners want to communicate with a real person. As he put it, “I personally never intend to have a meeting with someone else’s agent.”

That raises a bigger question. If AI agents become a normal part of everyday life, can businesses, websites, and payment systems handle the extra demand without becoming overwhelmed?

Embiricos said OpenAI is working closely with other companies as it expands what its agents can do. “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 said. He also mentioned ongoing discussions with payment processors, though he declined to share more details.

Allowing an AI assistant to make payments through partnerships with payment providers could open up new possibilities for OpenAI. An assistant might eventually handle purchases or pay for services on a user’s behalf. But giving AI that kind of financial authority is a significant step.

Allowing an assistant to read an email inbox already requires trust. Giving it access to a bank account is an entirely different matter.

For now, the promise of always-on AI assistants comes with some important questions about privacy, control, reliability, and how these systems will fit into the world around us. The technology is moving forward, but earning users’ trust may prove just as important as making the assistants smarter.