OpenAI Gave AI Its Own Computer. Can You Finally Log Off? (Dots)
OpenAI says Dots are AI agents that keep working when you aren't watching. They sit on their own cloud computer, use a browser, connect to apps you allow, and supposedly build up context about your preferences over time. The pitch sounds great. The reality, we will see.
On this channel we have been building with Astra, Codecs, and Higsfield. So the question is not whether Dots can complete a single task. It is whether handing one a job actually saves time. If I give an AI a goal and then spend an hour answering questions, checking every step, and fixing the results, I have gained a second job, not an assistant.
The test we want to run
I have a simple creator workflow in mind, the kind OpenAI itself uses as an example. I would give a Dot a transcript with timestamps and ask for three clip candidates. For each one, it needs to show the exact section, explain why it works on its own, and tell a viewer what they would learn. Then I want show notes and two social drafts written in my own style. Nothing gets posted, everything stays in draft.
The fun part comes next. Halfway through, I change the brief: make it for somebody who has never used an AI coding tool, explain the jargon, and remove any line that oversells what the footage actually shows. That is where a useful agent shows its worth. Does the correction land in every deliverable, or does the post get fixed while the show notes still say the old thing? Does it pick a clip because the line sounds exciting even though the next sentence flips the meaning? If the transcript does not support a claim, does it flag the gap or just fill it in confidently?
Either version is possible right now. The good one means I come back to organized drafts, correct references, and a few editorial decisions. The bad one means 20 notifications, six questions I thought I already answered, and a pretty document that still needs a rewrite.
What 'always on' actually means
OpenAI separates proactive research from the actions a Dot takes afterward. The safety documentation says proactive research only uses read-only tools on permitted connected sources. Those research tasks cannot message other people, change app content, or control a browser or desktop on their own. Follow-up actions still go through the normal rules and checks.
There is also a separate system called AutoReview that checks relevant actions against your instructions and safety requirements. So finding something that might need attention does not automatically mean the Dot has permission to act on it. That is important because the always-on label makes it sound like Dots are just out there doing things. They are not.
For a first test, I would keep the scope small. One transcript, a few clearly defined outputs, and an obvious handoff point back to me. I would watch the quality of its questions too. A useful question helps me make a decision. A pointless one just hands the task back.
Who gets access and what it costs
Dots are rolling out to Pro and Business Premium users in eligible markets. Enterprise users can try the beta if their workspace admin enables it. Your first Dot is included at no extra cost on Pro or Business Premium, but deeper work has an allowance. Do not read always on as unlimited work.
Setup starts on desktop, and access to your own computer is optional and off by default. Rollout is gradual, so check your own account before assuming you have it. I have not verified a full regional list, so I will not promise access in any specific country.
The verdict waits for the work
The interesting part of Dots is not picking a clip. It is follow-through. Understanding why I rejected something, updating the rest of the package, and not making the same mistake on the next video. That is what I want to measure when I get access. How much of the work is actually usable? How many times do I have to step in? And after all the checking and fixing, did I save any time?
A folder full of output does not mean the job is done. It might just mean the checking is about to start. I am interested in trying Dots, but the verdict has to wait for the work.