The problem with most productivity advice is that it gives you another system to maintain. Another list. Another dashboard. Another place to explain what you already know.
OpenAI's dots are interesting because they are designed to keep work moving between conversations. You give your dot a responsibility, keep talking to it as the situation changes, and review the work it brings back. The question is whether you can turn that into something useful in your actual week.
Here is a practical way to start: choose one unfinished project, give it a concrete deliverable and make the first result easy to inspect.
What is an OpenAI dot?
A dot is OpenAI's always-on agent in ChatGPT, powered by GPT-6 Astra. It has its own cloud computer and browser, can work across connected tools, and can continue cloud work while your computer is off. It can research, analyze data, prepare documents and build software. See OpenAI's introduction to dots.
That does not mean every task can run with your laptop shut. Work that uses your connected local computer needs that device online with the ChatGPT app open. Connecting a messaging channel also does not automatically connect your inbox, files or computer. These are separate connections. OpenAI's setup guide explains the distinction.
Check access, then create your dot
As of October 3, 2026, dots are rolling out gradually. OpenAI lists eligible Pro 100, Pro 200 and Pro 500 accounts for adults outside the EEA, UK and Switzerland, plus Business Premium and Enterprise rollout worldwide. Enterprise administrators must enable dots. An eligible plan does not guarantee that your account already has access. Check the current access requirements.
Open dots in ChatGPT on a desktop browser or in the desktop app and follow setup. You can connect apps then or add them later. Local computer access is optional. Create your dot on desktop first; the supporting mobile app update lets you continue the conversation, while mobile web is not supported. Follow the official getting-started steps.
Name it if you like. Then give it some work. Choosing the eyes and glasses is the easy bit; explaining what a finished job looks like is what makes it useful.
Delegate a responsibility with a visible finish line
Try a job with enough moving parts to make coordination useful, but few enough that you can check the outcome. Preparing a client meeting is a good example. The sources are limited, the deadline is obvious and you know what you need when you walk into the room.
Help me prepare for Thursday's client meeting. Use the proposal, the latest email thread and the notes I attach. Produce a one-page briefing with agreed scope, open questions, decisions required and next actions. Link important facts to the source. Draft a meeting agenda. Do not send messages or change calendar entries. You are finished when both documents are ready for my review.
This gives your dot a job, not a vague wish. It also prevents a common failure: receiving a long explanation of how to prepare for a meeting when you asked for the briefing itself.
The examples in this article are proposed assignments. The apps and actions available to your dot depend on your account, connections and permissions. If it reports missing access, narrow the assignment or provide the relevant files through a supported connection.
Give it the context that changes the answer
Context is not the same as a huge pile of files. For a client briefing, the signed scope matters more than an old sales deck. The latest confirmed date matters more than the first date discussed. Tell your dot which source wins if two documents disagree.
Add three short instructions to your brief: which material is authoritative, which information is private, and which tradeoff matters most. A launch plan optimized for the earliest date looks different from one optimized for the least disruption.
The signed proposal is the source of truth for scope and pricing. Use later emails only for confirmed scheduling changes. Keep client information inside this task. If the sources conflict, show me the conflict and continue with the parts you can verify.
You can steer the work while it runs. If the meeting moves or the client changes priorities, add the correction to the conversation and ask your dot to update the deliverable. You do not need to start from scratch every time a detail changes.
Three jobs worth trying first
1. Build a decision brief
Ask for options against your actual constraints. If you are considering a new booking system, specify required integrations, the budget range and the records you must retain. Ask for dated source links and unanswered questions.
Compare three booking systems against these requirements. Separate confirmed features from assumptions. Include the published price and any usage limit that affects us. Recommend what to trial first and explain the tradeoff. Do not sign up, start a paid trial or contact vendors.
The value is a decision you can make. A list of twenty products gives you another research task. A shortlist with exclusions, evidence and a trial plan removes one.
2. Prepare a deliverable from messy notes
Use a small set of meeting notes, approved figures and an example of the format you want. Ask for an editable artifact and a short list of facts that need your attention.
Turn these workshop notes into a client-ready follow-up document. Use the attached template. Separate decisions, actions and open questions. Keep the owners and deadlines exactly as written. Do not invent an owner for an unassigned task. Deliver an editable document and list anything that needs my decision.
Check the artifact itself, including layout and links. A completed conversation is not proof that the right file exists or that its contents are correct. OpenAI explicitly recommends reviewing outputs and reported errors after a task completes. See tasks and memory.
3. Keep a recurring responsibility moving
Once the first result is useful, give the job a cadence. Define what changes are worth interrupting you for. A weekly status brief should make missed deadlines visible without flooding your day with routine updates.
Every Monday at 8:30 a.m. America/Toronto, review this project tracker and prepare a short status brief. Highlight deadlines at risk, decisions waiting on me and changes since the previous review. Keep unchanged items quiet. Deliver it in this conversation. Ask me before editing the tracker or messaging anyone.
For fixed recurring work, confirm that your dot has saved the schedule with the right timing and destination. A sentence about doing something next week is not enough evidence that a schedule exists. OpenAI's documentation distinguishes saved recurring schedules from the dot's ability to wake up and continue assigned work. Read how recurring tasks work.
Permissions: make the next action unambiguous
Drafting and sending are different responsibilities. So are researching a supplier and buying from one. Say which action you want, what it applies to and when it can happen.
OpenAI uses app permissions, built-in safeguards and automatic action review. Custom rules can add ongoing boundaries, but do not override those requirements or grant access to an app. A rule is not a security guarantee; the documentation says your dot can make mistakes. Read the controls guide.
For your first pilot, keep external actions behind review. When a draft is ready, check the actual recipient, attachments and content before authorizing a send. If you later allow a repeated action, keep the scope narrow enough that both you and the agent can identify when it falls outside the agreement.
In the desktop app, open your dot's profile and select Activity to inspect tasks, files, results and requests for input. You can change direction as the work continues. If a task needs a login or a decision, resolve that exact request rather than assuming the agent has stalled. OpenAI describes Activity here.
What should you measure?
Measure usable output, not the number of tasks started. How much time did you spend briefing the dot? How much review was necessary? Did the work reach the finish line you named? Was there an external action you did not intend?
For a client meeting pilot, a useful result means you can walk into the meeting with a correct briefing. For a recurring tracker review, it means a deadline risk reaches you early enough to act. Pick one measure that matches the responsibility and check it for a week.
If the output is vague, tighten the deliverable. If facts are wrong, tighten the source rules. If the agent keeps asking about routine details, spell out the ordinary choices it can make. If a decision affects your client, money or reputation, keep yourself in the loop.
Common questions about OpenAI dots
Can my dot work while my computer is off?
Cloud work can continue without your device. Tasks that depend on your connected local computer need it online with the ChatGPT app open. Identify where the task runs before relying on unattended work.
Do dots have unlimited usage?
No. Tasks a dot starts or manages in Work and Codex count toward those products' limits. Introductory allowances and dot conversation rules can change. Check your account's current limits and OpenAI's current pricing terms.
What is the best first task?
Choose work you already know how to judge: a meeting briefing, a supplier comparison or a follow-up document. Supply a small, current source set and name the deliverable. Expand after you can trust your review process.
For help building that habit with your team, explore AI training or AI coaching. For a custom process with defined integrations, see AI agents. You can also compare Meta Muse and read our OpenAI and Anthropic model guide.
Product details checked October 3, 2026. These workflows are editorial examples, not claims of measured dot performance. Header image: AI-generated editorial photography, not an OpenAI product screenshot.
