ChatGPT Dots vs Claude Cowork: Work Agents Head-to-Head

ChatGPT Dots and Claude Cowork are the two most serious answers to the same question: what does an AI that does office work actually look like? OpenAI's Dots, announced September 29, 2026, are always-on agents with their own cloud computers that follow you across chat, phone, and team tools. Anthropic's Cowork is a workspace where you hand Claude a task with files and connections, and it finishes the job, even from your phone with the laptop closed.

They overlap enough that buyers confuse them, and they differ enough that the wrong pick wastes a quarter. This comparison puts them head to head on home base, integrations, approvals, controls, pricing, and verdicts by company profile.

Key takeaways

• Dots live across channels with their own cloud computers; Cowork lives in a task workspace with your files, connectors, and browser.
• Dots connect to 4,000+ apps and reach Slack, Teams, and voice; Cowork connects files, Slack, Google Drive, and the browser through plugins.
• Both ship real approval systems: Dots with allow, block, and require-approval tiers plus Activity View; Cowork with per-task approvals and admin-controlled automatic mode.
• Dots roll out inside ChatGPT Pro and Business Premium with enterprise in admin-gated beta; Cowork is generally available inside paid Claude plans.
• The deciding factor is your team's habitat: messages and monitoring favor Dots, documents and analysis favor Cowork.

The core difference in one paragraph

A Dot is like a colleague who never leaves: it has its own isolated cloud computer and browser, keeps context as you move between ChatGPT, mobile, Slack, and Teams, takes voice calls, and keeps monitoring your goals in the background, doing read-only research freely while pausing for approval on actions you designate. Cowork is like a workbench you hand jobs to: you give Claude files, folders, connectors, and a task, it works through steps including browser actions, and cloud execution means a job you start on your phone keeps running with the laptop closed. One is built for continuity across the day; the other is built for completion of the task in front of you.

Head-to-head table

If you want each side explored on its own terms first, read our ChatGPT Dots explainer and the Claude Cowork small-business guide before continuing.

Integrations: breadth vs depth

Dots win on breadth. More than 4,000 app connections through OpenAI's plugin ecosystem means the agent can plausibly reach the odd tools in your stack, not just the famous ones, and the cross-channel design means the same agent continues a thread that starts in ChatGPT and moves to Slack. For teams whose work arrives as messages, mentions, and meeting follow-ups, that continuity is the feature that matters most.

Cowork wins on depth with files. Local files and folders are first-class citizens, browser actions cover the portals and web tools that have no API, and plugins package a repeatable job, its skills, its connectors, its sub-agents, into something the team can reuse. For teams whose work arrives as documents, spreadsheets, and folders, that depth matters more than channel coverage.

Evaluate integrations against your actual stack, not the count. List the five systems a pilot workflow touches, then check each side for those five specifically. A 4,000-app catalog that misses your industry CRM loses to a narrower connector set that covers it, and a file-native workspace that cannot reach your line-of-business app loses the same way. Breadth numbers are marketing until they intersect your list.

Approvals and admin controls

Both vendors learned the lesson that autonomy without approvals does not survive contact with real businesses. Dots let background research run read-only while individual actions can be allowed, blocked, or set to require approval; the Activity View shows ongoing work and lets you redirect it; sensitive jobs such as changing a password always stay with the user; credentials can be used without exposing passwords to the model; and the system can stop a Dot if it detects malicious instructions. Cowork approves per task, offers an automatically-approve mode under admin control, and is itself merging into the main Claude experience, where Claude decides whether a request is a quick answer or a full task.

At the enterprise tier the stories rhyme. OpenAI's specialist Dots for procurement, invoice processing, customer support, and commercial contracts get their own identities and selected system access, and OpenAI is working with Microsoft to bring them under Agent 365 controls. Cowork's enterprise story runs through role-based access, spend limits, and OpenTelemetry streaming of activity into monitoring systems, plus a five-level maturity model for staged adoption. Our agent pricing comparison covers what the governance layer costs across vendors.

For a small firm, reduce all of this to non-negotiables: approvals on for external sends and spending, read-only defaults for background work, separate credentials for the agent, and a weekly log review with a named owner. Either product configured this way beats either product left on defaults. The internal assistant service page shows how these controls look in a deployed setup.

Pricing and access

Dots are rolling out to ChatGPT Pro and Business Premium in eligible markets, with the first Dot included and extended limits during the first month, and Dot usage not counting toward plan limits initially. Enterprise, education, and healthcare get a beta that an admin must enable, off by default. Note the regional exception: Pro is unavailable in the EEA, Switzerland, and the UK at launch, while Business Premium covers all supported regions. Cowork ships inside paid Claude plans, Pro, Max, Team, and Enterprise, with the admin controls living in the team tiers.

Neither announcement gives a tidy per-seat total for a small firm, and launch-month terms like "extended limits during the first month" will evolve into permanent price schedules. Price the pilot, not the press release: one workflow, one month, seats plus metered usage, recorded. Revisit the math quarterly while both products are this new.

Verdicts by company profile

Two closing cautions. First, both products are young in their current form, and today's gap may not be next quarter's; prefer month-to-month commitments while you learn. Second, neither replaces customer-facing service automation. When the job is answering calls, booking jobs, and chasing leads, that belongs on purpose-built lead follow-up rails, not on a generalist work agent.

FAQ

What is the main difference between ChatGPT Dots and Claude Cowork?

Dots are persistent agents with their own cloud computers that monitor goals and follow you across ChatGPT, Slack, Teams, and voice. Cowork is a task workspace for finishing defined jobs with files, connectors, and browser actions, including cloud execution. Dots watch and continue; Cowork sits down and finishes.

Which integrations do Dots and Cowork support?

Dots connect to more than 4,000 apps through OpenAI's plugin ecosystem and reach users across ChatGPT, mobile, Slack, and Microsoft Teams. Cowork works with local files and folders, connector apps such as Slack and Google Drive, and browser actions, with plugins bundling skills, connectors, and sub-agents for repeatable work.

How do approvals compare between Dots and Cowork?

Both require human approval design. Dots run background research read-only, let actions be allowed, blocked, or set to require approval, and provide an Activity View for live supervision. Cowork approves per task with an admin-controlled automatic mode. In both, sensitive actions like spending money or contacting customers should require a named human.

Which should a small business pick, Dots or Cowork?

Message-heavy teams that live in Slack or Teams and need monitoring plus follow-through lean toward Dots. Document-heavy teams producing proposals, reports, and analyses lean toward Cowork. Firms under ten people should pilot one workflow on one side for a month and let measured hours saved decide.

Next step

List the five systems your pilot workflow touches, check both sides against that list, then trial the winner for one month with approvals on and usage tracked. If you want help running that bake-off and wiring the winner into your operations, book a call or start with the free six-step AI automation plan on our homepage.