The AI That Keeps Working When You Stop

This week, OpenAI launched something different. Not a smarter chatbot — an agent that keeps working after you close the tab. They call it a Dot. It runs on its own cloud computer, stays connected to your apps, and picks up tasks on its own. It only asks you before anything important happens.

That sounds like a small upgrade. It is actually a different idea. Until now, AI was a tool you reached for. You opened it, typed something, got an answer, and moved on. A Dot is more like a background worker — it notices things and gets started without waiting. One early user found their Dot had spotted an unpaid invoice, written a draft, and was waiting for their go-ahead.

The setup matters. Dots connect to over 4,000 apps — email, Slack, Teams, your documents, your code. They run on GPT-6 Astra, OpenAI's most capable model. There is also a cheaper version, GPT-6.1 Sol, which cuts repeated-task costs by about 95% — cached text costs just $0.10 per million tokens. The first Dot is free with a Pro or Business Premium plan.

The real challenge is not the technology. It is trust and permissions. An agent that can send emails, submit code, and draft proposals needs clear rules. OpenAI added approval gates and blocked certain actions by default. But your team still has to decide where the line sits — which tasks it can own and which ones need a human sign-off. That is a management question, not a technical one.

The practical move: pick one specific, repetitive task and run a Dot on it. Not to save hours, but to learn the new skill — supervising an agent instead of prompting one. Teams that figure this out early will have a clear head start.