OpenAI Dots review: always-on agents with a cloud computer

Who this is for

You are deciding whether to turn on OpenAI's Dots for a team that already pays for ChatGPT Pro, Business Premium, or Enterprise. You want background work done—drafts, research, small fixes, follow-ups—without opening a new thread for every step. You also need a clear answer on what still requires a person.

The capabilities below are as OpenAI published them on September 29, 2026, and current as of October 12, 2026. Verify each one on one of your own workflows before you widen access.

What Dots claim to do

OpenAI describes Dots as always-on agents powered by GPT-6 Astra, each with its own cloud computer, browser, and learning from feedback over time. Through an ecosystem of plugins, they can connect to over 4,000 apps. You reach a Dot in ChatGPT (desktop, web, mobile), and OpenAI says they also work in Slack and Teams, with texting planned later.

Dots can take projects and run while you work on something else. OpenAI's examples include drafting pull requests from recurring feedback, revising launch materials when scope changes, updating analysis when new evidence arrives, refreshing a sales proposal as requirements shift, and turning interview transcripts into clips and posts for approval. An early-tester anecdote in the launch post: a Dot noticed a missed invoice, prepared it, and sent it after approval.

You can open the Dot's computer to inspect its work. OpenAI also describes permission to connect a Dot to your laptop when you choose. Built-in rules decide when to act, when to ask, and when to leave sensitive tasks (such as changing a password) with you. Custom Rules can allow, require approval, or block specific actions. Auto-review checks actions that could affect accounts or share information. Proactive research in the background is described as read-only on connected apps until you are actively working with the Dot.

Specialist dots for organizations are a separate track: company-provisioned identity, credentials, and access to systems of record, starting with focused enterprise pilots, including work with Microsoft Agent 365 for governance. That is not the same product as the personal primary Dot shipping to Pro and Business Premium.

Pricing as published

As of the September 29, 2026 launch post: Dots are rolling out to Pro and Business Premium users in eligible markets. Enterprise (including Edu and Healthcare) can try a beta when a workspace admin enables it. The first Dot is included in Pro or Business Premium at no extra cost, with an allowance for deeper work and extended limits for the first month after launch. Conversations with your Dot do not count toward ChatGPT usage limits; tasks started in Codex or ChatGPT Work still count as usual. OpenAI says you will later be able to add more Dots and scale speed or monthly work volume. Exact deeper-work allowances, per-market eligibility, and future add-on prices are not fully itemized on that page—check your plan's current billing screens and Help Center before you budget.

Where it falls short

Always-on and proactive sounds powerful; it also moves risk into the background. OpenAI itself says Dots can still make mistakes and that consequential work should be reviewed. The launch post does not publish completion rates, average time-to-approval, or failure modes by task type. Plugin access to thousands of apps is only useful if your team's permissions, data classification, and audit needs are already clear—otherwise you inherit every connected app's blast radius.

Specialist dots and deep systems-of-record integrations are still in pilot language, not a self-serve checklist. If your real need is an agent that owns one job inside your stack—with its own identity, deny shapes, and a handback packet when stuck—Dots' personal primary Dot may sit next to that need rather than replace it. Our notes on giving an agent a job, not a personality and naming what the agent cannot reach are the operating questions Dots do not answer for you.

Business and Enterprise data-use defaults differ from personal plans; read the workspace settings before you assume training controls match your policy.

A first-week pilot outline

If you enable a primary Dot on an eligible plan, keep the first week small:

• One owner, one workflow (for example: draft weekly status from linked docs for human edit).
• Plugins limited to read-mostly tools until Custom Rules for send/spend are set.
• Daily glance at Activity View; one intentional denial to see the approval path.
• A written note: what the Dot did well, what it invented, what you will not delegate yet.
• No specialist-dot conversations until that note exists.

That pilot is how you separate launch prose from your operating reality. It also creates evidence for IT and finance if someone asks why ChatGPT suddenly needs more connected apps.

Remember OpenAI's own caution in the launch post: Dots can still make mistakes; review consequential work. Your pilot should make "consequential" concrete—customer email, production code, money, legal commitments—not leave it as a slogan.

Alternatives and a plain verdict

Alternatives to weigh, depending on the job: a narrowly scoped custom agent on your own infrastructure (when containment and audit are the product); Microsoft-centric agent governance if you are already all-in on that stack; or simply better ChatGPT projects and human-owned checklists if background autonomy is not worth the permission surface yet. Consumer messaging agents such as Tab (reviewed separately) solve a different problem—personal errands over iMessage/WhatsApp—not team software delivery.

Verdict (as of October 12, 2026): Turn on a primary Dot for a small pilot if you already pay for an eligible ChatGPT plan and can name one workflow with clear approvals. Do not treat Dots as a substitute for a job-scoped production agent until you have watched Activity View, Custom Rules, and one failed run end to end. An always-on agent is only as useful as the approvals and records you put around it.

We design and build AI applications and agents with those boundaries in mind. If you want help picking a first Dot workflow—or deciding Dots are the wrong layer—talk to us.

Related work: Spin:Market.