OpenAI Dots vs Copilot Cowork: Which Fits Your Team?

OpenAI dots vs Copilot Cowork is a workplace decision about accounts, shared files and responsibility. Both products can carry out multi-step work. The practical question is whether your assignment belongs in a ChatGPT workspace or in the Microsoft environment where your colleagues already collaborate.

This comparison uses public official documentation checked on 30 September 2026. Microsoft’s Copilot Cowork is separate from Anthropic’s Claude Cowork; dots vs Claude Cowork covers that other product. A shared word in the name does not establish identical access, billing or permissions.

Start with the workplace you actually use

Microsoft’s current Cowork overview lists work across Outlook, Teams, OneDrive, SharePoint and Office documents. That makes it relevant when the material already lives in your organisation’s Microsoft account. Ask whether the proposed work can use the correct shared file rather than creating another copy that someone must reconcile later.

OpenAI’s dots introduction describes agents for continuing responsibilities, using computers and connected apps. The distinctive decision is whether you want to hand over an ongoing assignment alongside the work you already do in ChatGPT. What OpenAI dots is explains that product without treating the underlying model as the whole experience.

Neither description proves that a product fits every department. A sales team’s meeting preparation, a finance team’s spreadsheet review and a developer’s release work have different sources and approval requirements. Start with the process you want to improve. Do not select an agent first and then invent a task to justify the subscription.

Access and billing need a real account check

Microsoft’s current overview distinguishes generally available work or school access from personal preview access. Its Cowork product page describes usage-based work. Check the administrator’s enablement and the billing arrangement for the account you will use. An old preview announcement is a poor substitute for current documentation.

OpenAI’s dots account guidance lists eligible plans and regional restrictions. An eligible subscription and a feature already visible in your workspace are different things during rollout. Ask your administrator to check the intended workspace before making a recurring business process depend on it.

We do not quote a universal per-task price. Usage, entitlements and paid services reached during a task can all affect the cost. For a trial, record the work completed, any retries and the account’s actual usage record. Compare useful completed outputs rather than subscription labels alone.

A reasonable procurement question is: “What happens when this allowance runs out?” The answer should cover whether work stops, whether someone must approve more spending and whether the owner can still retrieve the output. It should also identify the person responsible for reviewing usage. An agent that keeps working needs someone who keeps watching its bill.

Shared Office files are a meaningful difference

Microsoft’s instructions for using Cowork describe editing an existing Word, Excel or PowerPoint file in OneDrive or SharePoint, with its existing version history. This is more specific than saying an agent “makes documents”. An edit to a shared workbook can affect people who never saw the prompt.

For that reason, distinguish a draft from an update to the source file. If you ask for a correction to the team’s forecast, specify whether the result should be a suggested change or an authorised edit. Include which sheet and which assumptions are in scope. Ask for an explanation of any formula or reference changed.

With dots, check the connection and the account used to reach the file. The computers and apps documentation separates cloud resources, local access and connected apps. Do not assume that a file visible in your personal browser is also available to the dot’s cloud computer.

The practical choice may be decided by ownership. A useful output in a personal account can still be unsuitable for a team workflow. Before either trial, agree where the source material belongs, where drafts should be stored and how colleagues will review the result. These are process decisions, not a test of which brand writes a nicer paragraph.

A worked example: preparing a customer review

Imagine an account manager preparing a quarterly customer review. The approved source material is a shared activity summary, a public customer website and a sample contract with confidential details removed. The required output is a briefing with achievements, unresolved issues and questions for the meeting. There is no permission to contact the customer.

Give the agent a written definition of the customer, the reporting period and the documents it may use. Ask it to identify the source behind each material claim. If it encounters a missing renewal date, it should list the uncertainty rather than infer a date from a previous meeting note. That uncertainty is useful preparation, not a defect to hide.

In a Microsoft trial, ask for a separate draft before allowing changes to a shared presentation. In a dots trial, ask for a briefing first and check which connected sources were used. Keep the comparison fair: the same source material, the same expected output and the same limit on actions. Different access makes a direct speed comparison misleading.

Review the draft with someone who knows the account. Did it distinguish a customer’s request from an agreed commitment? Did it include a problem that was already resolved? Did it attach the right reporting period to a number? Check these details before judging the prose. Checking an AI answer provides a useful method when another colleague holds the subject knowledge.

Then introduce a revision. The meeting now focuses on an unresolved delivery issue. Ask for a revised agenda and a list of statements that need human confirmation. Assess whether the agent preserves the facts while changing the emphasis. This is a proposed trial you can run; we have not measured either product on this assignment.

Approvals can be broader than a single click

Microsoft’s action guidance includes options to approve once or skip similar approval prompts for the rest of a session. That matters when setting up a trial. Read the choice you are making: permission to send one reviewed message is different from allowing later messages under a broader session rule.

OpenAI’s dots controls describe permissions and action review. Custom rules express instructions but can be followed incorrectly. Dots permissions and approvals explains the practical distinction between access to information and permission to act on it.

Keep the first assignment narrow enough to inspect. If the task later needs a customer email, review the recipient, message and attachments together. A document with another customer’s information is still wrong even if the email itself is well written. Ask for a fresh review when the task changes materially.

Recurring work needs an owner and an end

Microsoft documents scheduled and event-driven tasks; OpenAI explains ongoing work in its tasks and memory guide. In either case, a useful recurring assignment identifies the source, trigger, destination and person responsible for exceptions. “Keep our customers happy” does not define a reliable workflow.

For the review example, a recurring task might draft a private summary of new issues for the account manager. Specify which account it covers and when it should stop. A draft for an employee and a message sent to a customer are separate destinations. Review the stored task after setup so the agent’s interpretation matches the intended workflow.

If the process itself is unclear, workflow AI is a better starting point than adding another agent. Map where information arrives and where someone makes a decision. Automation becomes easier to judge when the existing handoff is visible.

Which one should your team trial?

Trial Copilot Cowork when the work belongs in Microsoft’s shared documents and communications, and your organisation has verified access and billing. Trial dots when continuing assignments alongside ChatGPT match the work and the right workspace can reach the approved sources. If neither account fits the data, fix that before proceeding.

Start with a bounded dots assignment or the equivalent Cowork draft, then assess accuracy, corrections, ownership and cost. Coursium’s approach to learning AI on your phone helps with the briefing and checking skills involved. It is a learning app rather than a substitute for your organisation’s agent platform.