Paul J Lipsky first treated ChatGPT Dots like a shopping or booking assistant and wondered what it added. It clicked when he gave it a different role: stay on top of work already happening across his ChatGPT projects and connected apps. In his hands-on video, a dot spots a hotel billing issue, reminds him about an unanswered email, and accepts three unrelated tasks in one voice call. The demonstration is useful precisely because it also shows a mistake in timing.
A dot is more than a fresh chat thread: OpenAI describes it as a persistent agent with a cloud computer, connected apps you permit, proactive read-only research, and task follow-through. Paul's best use is a chief-of-staff loop: notice, verify, coordinate, ask for decisions, and report back. His video shows promising alerts and task handoffs; it does not establish that every delegated job finished correctly or that Dots is the best agent for everyone.
Watch Paul's Dots Test
Credit: Paul J Lipsky's video, published 4 October 2026, and its supplied transcript are the source for the creator examples. Sponsorship: the video is sponsored by Outskill's GPT-6 Astra training offer; the sponsor segment begins at 4:20. The offer is not evidence for Dots' performance, and workshop dates or pricing should be checked on the sponsor page. Product availability and controls below are checked against OpenAI's documentation as of 5 October 2026.
Why Dots Clicked: Coordination, Not Another Chat
At 1:34, Paul contrasts a general personal-assistant job with his later discovery: Dots is useful when it can relate a request to other work already underway. OpenAI says a dot can handle several projects on its own computer, use connected apps, and start or manage tasks in ChatGPT Work and Codex. It can keep context across ChatGPT and supported messaging channels, but it can access only the apps and devices the user authorizes. A cloud computer is not automatic permission to read the local laptop.
That distinction changes the instruction. “Find me a product” is a one-off task. “Keep the Tuesday video moving, tell me when research is ready, and surface the decisions only I can make” is an ongoing responsibility. The second one needs status, evidence, and a definition of done. Without those, a chief-of-staff metaphor can hide unfinished work behind reassuring updates.
The Alerts: Useful, Then Too Late
Just after setup, Paul's dot appears to inspect connected context and flags three things around 2:26: a hotel billing discrepancy to settle before checkout, a video shoot that might be premature, and an email awaiting a reply. These are creator-observed suggestions, not proof that the agent checked every source correctly. They are also the kind of small, time-sensitive omissions a human chief of staff might notice.
The sharper lesson arrives with a sponsor payment. Paul asks the dot to send an email because he believes an invoice is unpaid. About half an hour after the email goes out, the dot finds evidence that payment may already have arrived. Paul confirms it and sends a correction. Proactivity helped recover, but a pre-send payment check would have prevented the awkward message. For anything involving money, reputation, or a customer relationship, make the workflow check the ledger or receipt, then draft, then approve. An agent's confidence is not the second source.
OpenAI says background proactive research is restricted to read-only tools. Sending the payment email was a separate requested action with its own review path. That boundary matters: a useful alert is not permission to send, buy, book, or edit on the user's behalf.
One Voice Call, Three Workstreams
At 5:48, Paul calls his dot and gives it three different outcomes:
- Research a ChatGPT Space video: gather accurate notes and shape an outline in an existing Obsidian scheduling page so Paul can write the final script.
- Improve Dayboard: in the original app version, open and lightly edit an Obsidian task page inside Dayboard while keeping an “open in Obsidian” path.
- Fix the game Schmela: smooth movement on steep hills and add unlimited arrows plus faster movement in a testing mode.
The dot repeats the brief and says the work is lined up. That demonstrates intake and coordination; the clip does not show a completed code review, passing tests, or a verified final outline for all three. Paul's interpretation at 9:47 is that one dot can continue following the threads and notify him when it needs a decision. For a real project, add an acceptance test to each task: the file path and sources for the outline, an editable-note test in Dayboard, and reproducible before/after steps for the game bugs.
Paul finds the voice-call presentation a little theatrical, but the ability to hand over several tasks without opening each project is practical. Do not treat “please avoid sensitive details in this call” as a privacy control; review what the dot can access, what is recorded, and what it actually says before using voice around confidential work.
Dots, Meta Muse, and Grok Bot
Paul prefers Muse for some everyday shopping or booking jobs and describes Grok Bot as a more independent team of agents. Those are his judgments from using the products, not a controlled speed, price, or accuracy comparison. The distinctive fit he found for Dots is coordination inside the ChatGPT work he already has. OpenAI's launch page says one primary dot is available first, while multiple or specialist dots are a future expansion or enterprise pilot; Grok Bot already documents multiple bots working together. Compare the task and permission model you need, not just the product label.
Set Up a Small Chief-of-Staff Test
- Check eligibility first. The OpenAI setup guide describes the current plans, regions, and gradual rollout. Create a dot in the desktop app or desktop web when the feature reaches your account; mobile cannot create one at launch.
- Connect only one low-risk source. Review the app's permissions. Existing ChatGPT context does not mean every external account is safe to connect. Local-computer access is separate and off by default.
- Give a standing goal and a filter. For example: “Each morning, summarize unresolved project deadlines from this authorized calendar. Alert me only if a decision is needed within 48 hours; otherwise include it in one digest.” Tell the dot which notifications are noise, as Paul did with ordinary incoming email alerts.
- Keep actions review-gated. Start with reading, comparison, and drafts. Require a person for external messages, bookings, purchases, code merges, and account changes. Check OpenAI's Custom Rules and approval model before broadening permissions.
- Inspect the work after a week. Review In progress, Scheduled, and Completed activity, then sample its sources and outcomes. Count useful alerts, false alarms, late discoveries, and tasks finished to the agreed standard.
One extra data caution: OpenAI says disconnecting a connected app stops new access, but does not erase information the dot already learned from it. Check the reset and deletion controls before connecting sensitive accounts.
Who Can Use It, and What Counts Toward Usage?
As of 5 October 2026, OpenAI lists a gradual Dots rollout for eligible Pro users outside the EEA, Switzerland, and UK, and for Business Premium users across supported ChatGPT regions. Enterprise, Edu, and Healthcare workspaces can try an admin-enabled beta. For a personal Pro account in Portugal, the regional restriction is a real blocker today; a promised wider rollout is not an access date. See the current eligibility page before changing plans.
OpenAI says the first dot is included with an eligible plan. Conversation with it does not consume normal ChatGPT usage limits, but deeper work has an allowance, and tasks it starts in ChatGPT Work or Codex count against those products' usual limits. Paul quotes a plan price in the video; because plans and regional prices can move, use current plan pricing rather than that quote to budget a pilot.
Turn the Chief-of-Staff Loop Into a Business Idea
The transferable skill is not “own a dot.” It is building a narrow loop that notices a real event, checks it before causing trouble, and brings the right decision to a person. This prompt works in any AI assistant.
Find a job worth staying on top of
Three buyer-specific offers, one read-only pilot.
Video Chapters
0:00 Intro · 0:33 What are Dots? · 1:34 Where Dots fits · 2:26 Proactive help · 4:20 Outskill sponsorship · 5:48 Multiple tasks · 9:47 Chief-of-staff role · 10:52 Using Dots · 11:47 What comes next.
Source Map
- Creator test: Paul J Lipsky's video and the supplied transcript. Its examples are account-specific observations; the three delegated tasks are not all visibly completed in the clip.
- Product facts: OpenAI's Dots announcement and Help Center setup, plans, permissions, and usage. For the comparison, see Grok Bot's announcement and Meta Muse.
- Sponsor: Outskill's linked GPT-6 Astra training offer. This article makes no independent endorsement of the workshop.