OpenAI is putting Dots in front of more people. The company is testing its personal agents on the Plus tier, which costs $20 per month. Dots are built to be proactive: they can work on a user’s behalf without supervision or assistance.
That is close to everything the public record confirms so far. We have a product name, a price point, and a pitch built around agents acting on their own. What we don’t have is a task list, usage limits, or a rollout schedule. So this review-section piece does something different from a hands-on verdict: it sorts what is known from what a prospective user should find out before relying on Dots for anything that matters.

What the announcement actually establishes
Start with the verbs. OpenAI is “expanding access,” and the Plus tier is being used for testing. Expanding implies that Dots was available to a narrower group before, though the details of that earlier group aren’t spelled out. Testing implies the feature isn’t final. Anyone paying $20 per month should treat this as an early look, not a finished product with a settled feature set.
The second thing established is the description of the agents themselves. Dots are called proactive, and they can operate without supervision or assistance. Those are strong claims, and they define the product more than any individual feature would. An assistant that waits for a prompt and an agent that acts on its own are different kinds of software, and they should be judged by different standards.
Why “proactive” changes what you are buying
A chatbot is reactive. You type, it answers, and you decide what to do with the answer. Every step has a human checkpoint built in, because the human is the one pressing send. If the output is wrong, the cost is usually a wasted minute and a rewritten prompt.
A proactive agent removes that checkpoint. If Dots can act without supervision, then the quality of its judgment matters in a way it never did for a chat window. A bad answer in a chat is an annoyance. A bad action taken while you are away from your screen is a different category of problem, and it may be one you only discover later.
That raises questions the announcement does not answer. What can a Dots agent touch? Can it read your accounts, or also change things in them? Does it ask before spending money, sending messages, or deleting anything? Autonomy is only as useful as the boundaries around it, and those boundaries are what separate a helpful agent from one you have to babysit anyway.
There is also a quieter issue: visibility. An agent that works unsupervised still has to report back somehow. A useful Dots experience probably depends less on what the agent can do and more on how clearly it shows what it already did. A clean activity log is worth more than another capability, because it lets you catch a mistake on the same day it happens.

The $20 question
Putting Dots on the $20 Plus tier matters because of who pays for it. Plus is not a niche enterprise plan. It is the plan individual subscribers buy for everyday use, which means testing here puts autonomous agents in front of ordinary users and not only specialists who expect rough edges.
Whether that is a good deal depends on facts OpenAI has not yet shared. If Dots runs under tight usage caps, the “works without assistance” promise may be limited by how often it can work. If the caps are generous, $20 for a standing agent could look inexpensive next to the time it saves. Until the limits are public, the price tells you where the feature sits, not what it is worth.
How to judge Dots if it shows up in your account
If you are in the test group, resist the urge to hand Dots something important on day one. Start with tasks where a mistake is cheap and easy to reverse. The point of the first week is to learn how the agent behaves when nobody is correcting it, and low-stakes errors teach that lesson at low cost.
Pay attention to three things. First, whether Dots does what you meant or only what you literally said, since unsupervised agents tend to fail in the gap between the two. Second, how it handles ambiguity: does it stop and ask, or push ahead on a guess? Third, whether it surfaces its own work clearly enough that you could audit a day of activity in a couple of minutes.
Also keep notes on the opposite failure. An agent that checks in constantly defeats the purpose of being proactive, and one that never checks in can quietly drift. The right balance will differ by person, and OpenAI is presumably running this test on Plus to see where most users land. So the practical question for anyone given access is a specific one: when Dots acts while you are not looking, would you be comfortable reading the full record of what it did?








