OpenAI's dots work 24/7. The rules you set matter more

The leak from last weekend turned out to be real, and it has a name. At DevDay on Tuesday, @OpenAI posted "Introducing dots, powered by GPT-6 Astra," calling them "always-on agents built to handle everything." About forty minutes later @sama followed up: "Dots are here!" He pitched them as something that "works 24/7 for you" so you get time and attention back.

Microsoft already has Copilot Autopilot and Meta has Muse, so OpenAI is joining a crowd of agents that keep running after you close the tab. For a small business, the question worth asking is what an agent like this may do while nobody is watching it.

What a dot actually is

According to OpenAI's announcement and early coverage from MarkTechPost and OpenTools:

• Each dot gets its own cloud computer with a browser, runs on GPT-6 Astra, and learns from your feedback over time.
• It connects to more than 4,000 apps through ChatGPT plugins, and you can talk to it in ChatGPT, Slack or Teams.
• When you are away it can do background research, but that mode is read-only: no sending messages, no editing content, no driving the browser.
• Anything that acts on your behalf goes through Custom Rules, where you mark an action as allowed, blocked or "ask me first." An auto-review step checks actions that touch accounts or share information, and sensitive tasks like password changes always stay with you.

It's rolling out now to ChatGPT Pro and Business Premium users in eligible markets. The first dot is included in those plans. Enterprise, Edu and Healthcare workspaces get a beta once an admin turns it on. OpenAI hasn't said what extra dots or heavier workloads will cost.

The plan lineup moved on the same day. Codex lead @thsottiaux laid out the new usage multipliers: Plus at 1X, Pro 100 at 5X and Pro 200 at 10X, with the $200 tier reopened to new subscribers. Anyone who was waiting on Pro now has the new numbers to weigh.

My read: the permissions matter most

Dots run on the model OpenAI already ships. The newer one, GPT-6.1 Astra, was pulled a day earlier after testing showed it pushing past its permissions (we covered that on Monday). So OpenAI launched its most autonomous consumer product on the older model and wrapped it in a lot of rules.

I think that was the right call. OpenAI itself says dots "can still make mistakes, so consequential work needs review." The read-only default for background work is sensible. An agent that researches overnight and hands you a draft in the morning is useful and low risk. An agent that sends the email at 2 a.m. without asking is how you end up apologizing to a customer.

Before trusting one with real work, I'd watch two things. The first is reliability over weeks. A demo shows one good run, but a background agent has to behave the same way on day 40, after a plugin has changed its API and the calendar has picked up a double-booking.

The second is permission creep. After a few good weeks, it's tempting to flip every "ask first" to "allowed." That approval screen is the only thing between the agent and your customers, so be stingy with it.

What this means for a local business

Most owners in the Fox Valley won't be on a $100 or $200 ChatGPT plan next week. The same thinking applies to whatever tool you end up using: an always-on agent is only as safe as the list of things you've decided it may do alone. Most businesses have never written that list down, even for their human staff.

Start there. Pick the jobs where a mistake costs almost nothing: summarizing reviews, pulling together a weekly numbers report, drafting follow-up texts that a person approves. Keep anything involving money, customer promises or account access on "ask first." Then look at which approvals you rubber-stamp every time, and loosen only those.

If you'd like help drawing that map for your own operation, New Face Design runs a free process audit. We go through where your hours actually go, flag what an agent could safely take over, and spell out what should stay with a person.