OpenAI's «dots»: A Personal Assistant That Never Stops Working

After years of talk about «chatting with AI», the question seems to have shifted: the focus is no longer on what a model can answer, but on what it can do on your behalf while you're busy with something else. This shift has produced a new category of products known as the «always-on» personal assistant — an assistant that keeps working in the background even when you haven't given it any instructions. OpenAI entered this space with a product launched on its Developer Day in late September under the name dots.

From a coding tool to a daily assistant

The most intriguing story behind this product isn't entirely technical. According to Alexander Embiricos, who leads the enterprise products team at OpenAI, most of dots' functionality can be achieved using Codex, the coding tool the company released earlier this year. The difference is that Codex demands more effort and constant follow-up from the user, whereas dots was designed to be intuitive and require no engineering expertise.

In other words, OpenAI didn't so much build a new technology as repackage its existing capabilities into a simpler experience. Embiricos says the team realized that framing the product as a single entity that talks to you in one continuous conversation — rather than dozens of scattered chats — is what makes it understandable and natural. Even the name is inspired by the empty space within OpenAI's logo, as something simple that can spread everywhere.

How does it work in practice?

The assistant is available within the ChatGPT desktop and mobile apps for subscribers on the Pro plan and above, with pricing starting at around $100 per month. Once connected to your work sources, it begins helping you get things done, including:

• Accessing your calendar, email, computer files, and connected tools.
• Flagging scheduling conflicts and upcoming deadlines.
• Taking the initiative to analyze situations and propose solutions without waiting for an explicit command.
• Running «proactive search», which reviews what it knows about you to suggest new ways it can help.

A striking example cited by Embiricos: an engineer at a drone delivery company told him that an outage had occurred, and before he even noticed it, his assistant had already detected it, analyzed it, and reached out with useful information to address it.

Proactivity is still imprecise

That said, the ability to «think ahead» doesn't appear fully mature. Alerts can arrive later than is actually useful. Embiricos recounts that the assistant once notified him of a scheduling conflict just three minutes before the meeting began, rendering the alert practically worthless. He expects the timing to improve gradually, but insists that proactive logic is the core of the product, not an add-on feature.

It's worth noting that this proactivity serves both sides: a richer experience for the user on one hand, and greater token consumption to the company's benefit on the other.

The inescapable privacy question

Every effective personal assistant needs a large volume of your personal data, and that's where the tension begins. Competing assistants have drawn criticism for building profiles of family and friends to personalize suggestions — data that could also serve advertising businesses. dots, according to OpenAI, relies on the same controls already available in ChatGPT: the ability to opt out of having your conversations used to train models, and the ability to define what the assistant can access on your device.

But these controls aren't yet granular enough; the app refuses, for example, to settle for read-only permission on cloud storage services. Embiricos acknowledges that what's needed is building «highly granular controls», while noting in return that the model's intelligence and good judgment are part of the equation. He mentions that the first release adopted very conservative settings and requested permissions frequently, which drew some criticism, but the team preferred to start from a safe zone and adjust from there.

What should you watch for?

The biggest question facing this category of products is scalability. An assistant that books a hard-to-get table seems magical when only a handful of people have one; but once it's in the hands of millions, the pressure on services, websites, and infrastructure changes dramatically. On top of that, the high monthly cost remains a barrier limiting its spread beyond professionals and enterprises.

The practical lesson for companies: the real value in this wave doesn't come solely from a more powerful model, but from redesigning the experience around the model and from clear data access policies. Whoever strikes the right balance between proactivity and privacy will win users' trust before winning their admiration.

✦ بقلم فريق دروب أيديا