OpenAI calls Dots an always-on personal agent. Futurepedia had early access and tried a more practical question: can one agent keep a video project moving across ChatGPT, Work, Codex, and a phone call without being told every next step? The answer in this first look is promising, but uneven. It coordinated several creative tasks; its cloud computer and a scheduled-task view also failed during the recording.
Watch Futurepedia Test Dots
Credit: Futurepedia's early-access video, published 29 September 2026. The creator says OpenAI supplied early access but did not sponsor the video. This article uses the supplied transcript, the Futurepedia and Skillap guide, and OpenAI's official product and help pages, checked 30 September 2026.
What Dots Is, According to OpenAI
OpenAI describes Dots as GPT-6 Astra-powered agents with their own cloud computer, persistent context, connected apps, and the ability to work toward ongoing goals. A dot can talk with you in ChatGPT, Slack, and Teams, and can take a voice call. It can initiate or manage work in ChatGPT Work and Codex. That is why the video frames it as a project manager across existing tools rather than a new chat box.
The cloud computer is not your laptop. OpenAI's setup guide says local computer access starts off and must be explicitly enabled in the desktop app. Connected apps also have separate permissions. For background proactive research, OpenAI says Dots uses read-only tools; that does not mean every action the dot may take in an active task is read-only.
The Official Demonstration
For the launch-day version of the story, compare Futurepedia's test with OpenAI's Dots segment in the DevDay keynote. A staged product demonstration shows the intended workflow; it does not cancel out the rough edges in an early user test.
What Futurepedia's Test Actually Shows
| Test | Observed result | What to take from it |
|---|---|---|
| Setup and controls | The creator renamed the dot, inspected its cloud computer, and briefly enabled local-computer access. | Local access is a deliberate choice. Review it separately from the dot's own computer and connected apps. |
| Project Pages | The dot made a scratchpad and a nested page for a Dots launch video, with notes, checklists, and links to prior chats. | Useful for keeping project context visible, but check every inherited note before reusing it. |
| Work + Codex + chat | It requested three intros in Work, selected one, sent it to a Codex motion-graphics skill, and started a thumbnail in another chat. | This is the strongest demonstration of coordination. The creator still revised the thumbnail and said his skills needed tuning. |
| Voice and browsing | A phone conversation found product links, but the cloud computer failed after a reset; a Codex browser provided a workaround. | Finding links is not the same as completing the requested browser action. The failure was in an early beta, not a measured outage rate. |
| Scheduling | A future keynote-summary task appeared, while the scheduled-task tab would not open in the demo. | The video shows task creation, not proof that the scheduled run later completed. |
The pattern matters more than any one polished asset: Dots remembered the larger video goal, delegated different pieces, and returned outputs for review. Futurepedia's judgment is that continuity is the differentiator. This single early-access session does not establish reliability, quality across users, or a guaranteed time saving.
The Creator's Free Dots Guide: Three Starter Jobs
Futurepedia and Saj from Skillap published a free Dots use-case guide on HubSpot. It explains setup, boundaries, and three copyable prompts. The full wording belongs in their guide; here is the practical choice behind each example:
| Responsibility | Give the dot | Keep with a person |
|---|---|---|
| Partnership and sponsorship triage | Typical rates, terms, deliverables, and the shape of a standard deal. | Approve replies and flag unusual exclusivity, usage rights, or payment terms. |
| Next content brief | Past scripts or posts, editorial standards, calendar, and topics worth watching. | Review sources and angle; approve anything published. |
| Known audience questions | An FAQ, help document, or other approved answer base. | Review unfamiliar questions and new answer policies; do not let the dot guess. |
The guide's best starting rule is one ongoing responsibility, not five. Set action rules before connecting sensitive tools and expand access only after you have reviewed real outputs. Its plan and feature notes may change; for current eligibility, use OpenAI's live Help Center.
Who Can Use It, and What Can It Do Without Approval?
As of OpenAI's 29 September launch, Dots is rolling out gradually to personal Pro accounts outside the EEA, Switzerland, and the UK. Business Premium is listed across supported ChatGPT regions; Enterprise, Edu, and Healthcare require an admin-enabled beta. That distinction matters to a reader in Portugal: a personal Pro subscription alone does not unlock the initial Dots rollout there. The first dot is included on eligible Pro and Business Premium plans, with a separate allowance for deeper work; Codex and Work tasks still use their normal limits.
A dot can carry context from memory and connected apps, so the permission decision is not just whether it may click a button. OpenAI says you can inspect its Activity View, pause it, disconnect apps, and define Custom Rules for actions. Its launch page says consequential actions may be subject to auto-review and approval, while certain sensitive actions stay with you. Review connected data and outputs anyway: an assistant can make mistakes even when it follows the rules.
A Careful First Test
- Choose one repeatable, low-risk job. For example, prepare a weekly content brief from public sources, or draft replies from an approved FAQ without sending them.
- Give it a definition of done. Specify the source set, format, deadline, and what must be escalated rather than guessed.
- Keep access narrow. Start with pasted or connected read-only information; leave local-computer and finance access off until the task needs them.
- Review the entire chain. Inspect sources, delegated Work or Codex tasks, final files, and any scheduled follow-up. Track corrections and failures across several runs.
If Dots is not available in your region or plan, the same experiment works manually in an assistant you already use. You can test whether the responsibility itself is valuable before paying for, or granting permissions to, a coordinating agent.
Futurepedia Video Chapters
| Time | Chapter | Time | Chapter |
|---|---|---|---|
| 00:00 | Introduction | 01:19 | Setup and configuration |
| 02:47 | Core capabilities and free guide | 04:03 | Working with Pages |
| 05:12 | Orchestrating Work, Codex, and chat | 06:10 | Voice and cloud-computer issue |
| 07:50 | Reviewing outputs | 09:28 | Scheduling |
| 10:19 | Final impressions |
Sources and Useful Links
- Hands-on test: Futurepedia's video and the transcript supplied for this article.
- Creator resource: Futurepedia and Skillap's three-use-case guide and full prompts.
- Official product: OpenAI's Dots announcement and setup, access, permissions, scheduling, and controls.
- Launch context: the official DevDay Dots demonstration and our full launch-status guide.