Editorial test of Dots reveals obstacles in online tasks and a need for human help
An editorial test of Dots from OpenAI encountered website security checks and a need for manual intervention. Editing the author's own website produced better results. The service is available on selected paid plans, with regional restrictions on access through the Pro plan.
An editorial test of Dots by The Verge revealed limitations in routine online tasks. When ordering internet service, the author had to complete verification manually, while other security checks blocked account login or food ordering. The agent also overlooked a free visit to a coworking space and offered paid entry. The author described a better experience when designing and repeatedly revising her own website using voice instructions. These are the results of a single editorial test.
OpenAI introduced Dots as continuously running agents powered by GPT-6 Astra. Each agent has its own cloud computer with a browser and, according to OpenAI, can connect to more than 4 000 apps. Users currently have one agent, which they can communicate with through ChatGPT, Slack or Microsoft Teams. They can grant access to their own computer through the ChatGPT desktop app; SMS communication is only planned at this stage.
Access is being rolled out gradually for Pro and Business Premium plans. According to The Decoder, the Pro plan does not yet provide access in the European Economic Area, Switzerland and the United Kingdom, while the Business Premium plan offers the service in all supported regions. For Enterprise, Edu and Healthcare plans, a beta is available once enabled by an administrator. The sources cast doubt on a consistent price figure: one article from The Verge lists the Pro plan at 20 USD per month, but the editorial test from the same outlet lists 100 USD per month. An actual price reduction is therefore not clearly established.
According to OpenAI, agents use only read-only tools when independently looking for opportunities to help. For active tasks, custom rules can require approval or prohibit certain actions; sensitive actions, such as changing a password, remain the responsibility of a human. The company acknowledges errors and recommends checking results that have real-world consequences. One agent is included in the subscription, and conversations with it do not count toward ChatGPT product limits, but tasks run in Codex and ChatGPT Work tools count toward limits in the usual way.
Why it matters
An agent working on its own cloud computer allows users to delegate tasks to run in the background, but security checks on third-party websites can interrupt their completion. The editorial experience of editing the author's own website shows a concrete use for iterative work based on instructions; however, the same test does not demonstrate reliable completion of routine online orders.
What was added since the original report
Verified updates
-
The editorial test recorded speech transcription errors.; The agent did not fulfill its promise to solve a CAPTCHA.; The agent addressed the author by the wrong name during a voice call.
- The editorial test recorded speech transcription errors.
- The agent did not fulfill its promise to solve a CAPTCHA.
- The agent addressed the author by the wrong name during a voice call.
-
Users currently have one Dots agent available.; The agent's virtual computer offers Blender and GIMP.; Users can grant the agent access to their own computer through the ChatGPT desktop app.; The agent needed human help when ordering internet service.
- Users currently have one Dots agent available.
- The agent's virtual computer offers Blender and GIMP.
- Users can grant the agent access to their own computer through the ChatGPT desktop app.
- The agent needed human help when ordering internet service.
-
The article lists the Pro plan price as 20 USD per month, compared with the previous 100 USD.; Dots take the form of colorful, customizable characters with eyes.; OpenAI demonstrated the ability of Dots to create virtual worlds.
- The article lists the Pro plan price as 20 USD per month, compared with the previous 100 USD.
- Dots take the form of colorful, customizable characters with eyes.
- OpenAI demonstrated the ability of Dots to create virtual worlds.
-
At launch, users can use only one Dot agent.; The author verified during his own testing that Dot suggests specific tasks based on ChatGPT history.
- At launch, users can use only one Dot agent.
- The author verified during his own testing that Dot suggests specific tasks based on ChatGPT history.
-
Dots runs on GPT-6 Astra; Available in ChatGPT for Pro and Business Premium users; Integrates with Slack, Teams and other organizational platforms; Launched on 29 September 2026 at the OpenAI Dev Day event; Agents work independently in the background without constant supervision
- Dots runs on GPT-6 Astra
- Available in ChatGPT for Pro and Business Premium users
- Integrates with Slack, Teams and other organizational platforms
- Launched on 29 September 2026 at the OpenAI Dev Day event
- Agents work independently in the background without constant supervision
-
Dots are available for $100/month for ChatGPT Pro users; They require explicit user consent for sensitive actions; They support Custom Rules for user-defined restrictions; Also available on iMessage and RCS (via a waitlist); They learn user preferences and scale to multiple agents at once
- Dots are available for $100/month for ChatGPT Pro users
- They require explicit user consent for sensitive actions
- They support Custom Rules for user-defined restrictions
- Also available on iMessage and RCS (via a waitlist)
- They learn user preferences and scale to multiple agents at once
-
Dots have access to more than 4000 supported apps; They have built-in and customizable safety rules; They are being rolled out gradually, starting with ChatGPT Pro users; Currently, only one Dot can be created (a future plan will allow more); They communicate via SMS, text, voice, Microsoft Teams and Slack, carrying context across channels
- Dots have access to more than 4000 supported apps
- They have built-in and customizable safety rules
- They are being rolled out gradually, starting with ChatGPT Pro users
- Currently, only one Dot can be created (a future plan will allow more)
- They communicate via SMS, text, voice, Microsoft Teams and Slack, carrying context across channels
-
Dots run continuously on their own cloud computers with a browser; They use read-only access in the background and require approval for sensitive actions; Users in the EU currently do not have access to the functionality; One Dot is included in a standard ChatGPT subscription, while business customers have expanded options; They autonomously handle specific tasks: research, data analysis, coding, document preparation
- Dots run continuously on their own cloud computers with a browser
- They use read-only access in the background and require approval for sensitive actions
- Users in the EU currently do not have access to the functionality
- One Dot is included in a standard ChatGPT subscription, while business customers have expanded options
- They autonomously handle specific tasks: research, data analysis, coding, document preparation
Two audiences, two different impacts
What this means
For individuals
For individuals, the ability to try the service depends on both their subscription and their region. Pro users in the European Economic Area, Switzerland and the United Kingdom do not yet have access.
For a business
When integrating agents into workflows, the company must plan for handing some steps over to employees and checking results. Automation may also count toward existing limits for Codex and ChatGPT Work tools, even though conversations with the agent itself do not count toward ChatGPT product limits.
ProcessesCheck the original
Event sources
confirmed by 4 independent sources · 5 publishers, 4 independent. We count feeds from the same owner only once.