OpenAI announced Dots at its DevDay event on 29 September, describing the product as an agent built to take on “ongoing responsibility” rather than respond to single prompts. Unlike ChatGPT or Codex, a Dot keeps working between conversations, running on its own cloud computer and adapting as circumstances change, according to OpenAI’s official DevDay recap and a companion safety explainer published the same day.

The launch lands as B2B SaaS marketing teams weigh how much of their workflow to hand to an agent that does not wait to be asked. OpenAI’s own disclosures show the answer is not “all of it”: Dots ships with a built-in approval layer that stops it acting alone on the kind of consequential steps a marketing team would want reviewed before they happen, from spending money to deleting a file.

What is Dots, and what can it do?

Dots is powered by GPT-6 Astra, which OpenAI calls its most capable and aligned model, built to understand a user’s goal, stay within the scope of a request and ask questions before acting on an ambiguous one. Each Dot gets its own cloud computer, where it can browse the web, analyse information, create files and run tools, and it can connect to whichever apps a user permits through existing ChatGPT connections. Users can also link a personal computer, so a Dot can work on a desktop spreadsheet or use locally installed coding tools.

OpenAI’s safety explainer gives two examples with direct marketing relevance: a Dot that follows a developing story and prepares a newsletter draft, and one that updates a project as new information arrives. The company does not claim Dots can run a campaign, nurture leads or write to a CRM, and nothing in its own material describes the product as working independent of every interface — OpenAI’s language is that a Dot connects to apps a user chooses, not that it operates without any platform at all.

Access is tiered. Dots is available now to ChatGPT Pro and Business Premium subscribers in eligible markets, while Enterprise, Edu and Healthcare workspaces can try a beta only if an administrator switches it on; it is off by default. OpenAI has not published a separate Dots price or usage allowance, though CNBC reported that access begins with the existing Pro plan. At the same event OpenAI introduced a new “Pro 500” tier offering 25 times the usage allowance of ChatGPT Plus, and GPT-6.1 Sol, a cheaper model OpenAI says delivers “near-Astra intelligence” at a fifth of Astra’s standard token price — neither is a Dots-specific change, but both reflect a broader push to get more agentic work in front of paying users.

What stops Dots from acting without approval?

OpenAI’s safety explainer is explicit that an agent mistake can have consequences outside a single chat, including changing the wrong file or sharing information that should have stayed private, and it names prompt injection, malicious instructions hidden in a webpage, email or document, as a live risk. Its response is a layered set of checks rather than a single switch. Users can set Custom Rules, such as telling a Dot never to send email on their behalf, and a separate Auto-review system checks every planned action against those rules and OpenAI’s own safety requirements before it runs.

Certain categories of action go further than a check: purchases need explicit approval, as do permanently deleting data, installing software from an unrecognised source and granting new security-sensitive access. Password changes and transfers between financial accounts are handed back to the user outright rather than attempted by the agent. OpenAI also says a Dot is not meant to infer open-ended permission to share information with a new recipient, and that disconnecting an app stops new data flowing in but does not erase what the Dot has already learned from it. For Business, Enterprise and Edu workspaces, OpenAI says content is not used to train its models by default; on personal plans, that depends on the account’s “Improve the model for everyone” setting. The company is careful to frame these as mitigations, stating plainly that Dots can still make mistakes.

How does Dots compare with Meta’s Muse?

OpenAI’s launch puts it in direct competition with Muse, Meta’s rival personal agent, which Reuters reported had already drawn millions of downloads earlier in September. Evercore ISI analyst Mark Mahaney has separately forecast that Muse could reach 100 million users within months, and told Investors’ Business Daily that Meta currently has the access advantage while OpenAI’s existing business relationships could give Dots an edge in higher-value professional work — a reported view, not a demonstrated outcome either way.

Neither company has published a reliable estimate of how far autonomous agents have actually spread into marketing workflows. A Digiday report found that 86% of brand and agency professionals’ companies were investing in AI in 2025, but that figure describes AI investment in general, not the use of always-on agents specifically, and should not be read as evidence that marketing teams are already running agentic work in production.

OpenAI has not said whether Dots will get its own price or usage allowance separate from the Pro and Business Premium plans it currently requires.