Tech Analysis
OpenAI's New Always-On Agents: Where Does the Work Actually Happen?
September 2026
At its DevDay on September 29, OpenAI introduced Dots: agents that stay on, keep working on a task after you've stopped talking to them, and do it on a computer of their own. Most of the coverage focuses on what they can do. The more useful question is a simpler one: where does the work actually happen, and what does that mean for you?
What's new
- Each dot gets its own cloud computer and browser, so work continues when your laptop is closed
- You reach a dot through ChatGPT, Slack or Teams once it's set up
- Plugins connect it to more than 4,000 apps
- It runs on GPT-6 Astra, and can do background research through read-only connections
- It's rolling out on Pro, Business Premium and Enterprise plans in eligible markets, and OpenAI's help documentation labels it beta
The controls OpenAI describes
The design puts a lot of weight on permissions, which is sensible for something that acts on its own:
- Default rules decide when a dot acts alone and when it asks for a go-ahead
- Custom rules let you allow, require approval for, or block specific actions
- An activity view, plus the ability to open the dot's cloud computer and watch it work
- Access to your own computer is optional and starts switched off
One detail from OpenAI's documentation is easy to miss: a dot's cloud computer doesn't inherit your VPN, your browser sign-ins or your device policies, so reaching some sites can mean signing in on that separate computer. Early coverage of the keynote also says that, for supported logins, a dot can use saved passwords without exposing them to the model.
Two ways to get AI working for you
Dots are one answer to "where does the work happen": on a cloud computer the provider runs. The other answer is on your own device. Neither is better in general. They suit different jobs.
Cloud agent (like Dots)
Keeps working when your devices are off, can reach thousands of apps, and can use the largest models. In return, you give it accounts and permissions, and the work happens on a computer you don't control. Your protection is the rules you set and the activity you review.
On-device AI
The data stays on your hardware, it works offline, and there are no accounts to hand over. In return, it's limited to what the device can do: it won't run tasks across your apps overnight, and the models are smaller.
Four questions before you hand a task to an agent
- Does it need my accounts, or can it work on a file I give it?
- What's the worst thing it could do, and does that action need my approval?
- Can I see what it did afterwards?
- Does it really need to run while my device is off?
If the answer to the last question is no, a tool that works on your own device is often enough, and it removes the first three worries.
What we don't know yet
This is based on OpenAI's announcement and early coverage. We haven't tried Dots ourselves, so we can't say how reliable they are day to day. It's also worth knowing the launch comes while OpenAI is still investigating earlier incidents involving its agents, including the Hugging Face breach in July. The controls described above look like a direct response to that, and real-world use will show how well they hold up.
Frequently asked questions
What are OpenAI's Dots?
Always-on agents announced at OpenAI's DevDay on September 29, 2026, powered by GPT-6 Astra. Each dot has its own cloud computer and browser and keeps working on a task after the conversation ends.
Who can use Dots?
According to coverage of the launch, they are rolling out in ChatGPT on Pro, Business Premium and Enterprise plans in eligible markets. OpenAI's help documentation labels the feature beta.
Can a dot control my own computer?
Only if you allow it. Access to your own computer is optional and starts switched off.
How is this different from AI that runs on my phone?
A dot does its work on a cloud computer run by OpenAI, so it can keep going when your devices are off and can reach many apps. On-device AI does its work on your own hardware, so the data stays put, but it is limited to what the device can do.
Try Offgrid Studio
Four iOS apps that put on-device AI to work for you.
More from Offgrid Studio
Also read: An AI Agent Broke Into a System on Its Own. Spain Just Logged It as a Data Breach. and I'm Not Afraid of AI. Twenty Years of Maintenance Taught Me Why.