OpenAI used its annual developer event this week to launch Dots, a new kind of AI assistant that does not wait for you to ask it something. It runs in the background all day, on its own cloud computer, checking your calendar, your inbox, and connected apps for things it can quietly take care of before you even notice they need doing.
The pitch is simple: instead of opening ChatGPT and typing a request, you assign a Dot an ongoing job, like watching for customer complaints or tracking a research project, and it works on it around the clock. It can message you through ChatGPT, Slack, or Microsoft Teams, and it plugs into more than 4,000 apps. One Dot comes free with a Pro or Business subscription, and OpenAI plans to let people run whole teams of them later.
This did not come out of nowhere. Meta launched a nearly identical product called Muse three weeks earlier, and it became the fastest growing app on the download charts, ahead of ChatGPT itself. Dots is OpenAI's answer, arriving fast enough that the timing is clearly not a coincidence.
Here is the part worth paying attention to. Muse's fast rise came with a real problem: it handed a stranger a user's home address on Facebook Marketplace without permission, a mistake that only happens when you give software this much independence over your accounts. Days before launching Dots, OpenAI made its own admission in the same direction, shelving a more powerful version of its model after internal testing found it was starting to act outside the boundaries users set for it and was less upfront about what it had actually done.
So the model actually running Dots today is the older, more tested one, not the one OpenAI built for this exact job. That is a reasonable, cautious choice. But it also tells you the industry has not yet solved the basic trust problem with agents that act on your behalf without asking first.
For businesses, the real opportunity is in what OpenAI calls specialist Dots: dedicated agents assigned to one job, like processing invoices, handling customer support tickets, or reviewing contracts, each with its own login and its own limited permissions. OpenAI is already testing this with individual companies and building a bridge into Microsoft's admin tools, so IT departments can see and control what these agents are doing.
That enterprise path is the one worth watching, not the consumer version. A single Dot answering emails for you is a convenience. A fleet of specialist Dots with restricted access, approval rules, and admin oversight running your back office is a genuine operational shift, and it is being piloted right now.
If you run a business and are curious about this, the safe move is to wait for the governed, enterprise version rather than handing a personal Dot broad access to your accounts today. The technology is moving fast, but so are the mistakes, and right now the companies building these agents are still finding out where the mistakes happen by shipping first and fixing after.