OpenAI's Dots Turns a Prompt Into a Persistent Worker. Marketing Teams Need Approval Rules.

OpenAI introduced Dots at DevDay on September 29, bringing the persistent-worker model that Grok Bot launched in August into ChatGPT.

A 5-to-30-person marketing team can now hand an agent a continuing job with context, rather than start over with a new prompt every time the brief, campaign or customer signal changes.

Dots are designed to keep working in the background from their own cloud computer. The Verge reports that they can use a browser and more than 4,000 connected apps, carry context into Slack and Microsoft Teams, and send updates or questions as they work. The launch is initially for ChatGPT Pro, Business Premium and Enterprise users, with one Dot per user.

“This is the real deal version of AI we’ve always imagined.”

Sam Altman, CEO of OpenAI, at DevDay, as reported by Axios.

That claim is bigger than a better chat window. A chat session finishes when the response appears. A persistent agent can keep the customer research, the approved positioning, the last campaign report and the next action in the same piece of work.

The marketing change is in the handoffs

OpenAI’s product demonstration included a creator workflow: review an interview transcript, identify clips, prepare show notes and draft social posts. The individual tasks aren’t new. Most marketing teams already use AI for one or two of them.

The new part is continuity. One agent can carry the same assignment through the steps that usually lose context: the transcript becomes an audience insight, the insight becomes a content brief, the brief becomes drafts, and the drafts return for approval with the source material still attached.

That is the lane Grok Bot established when xAI launched it on August 11. xAI described Bots as always-on teammates with their own computer that can work across tools, inboxes and apps. It listed marketing campaigns and sales outbound among the use cases.

The market question is whether a marketing team can turn a pile of one-off AI tasks into a system that remembers the work and respects the rules around it. Meta’s earlier Muse launch showed the same shift on the customer side. Dots brings it inside the team’s own work.

This story is distinct from our earlier look at Grok Bot’s public company-building experiment. That was a chance to watch a team work across product, sales and marketing in public. Dots makes the same category of long-running work available through the ChatGPT stack marketers already know.

Give the agent a lane before you give it access

A persistent agent can create a faster version of an existing workflow. It can also create a faster route around the people who own customer data, brand claims and publishing permissions.

Reuters reports that Dots lets users set custom action rules and requires explicit consent for sensitive actions such as changing passwords or permanently deleting data. The company says business data isn’t used for training by default.

Marketing teams should translate that into operating rules before they start delegating work. A Dot can research a target account, summarize a call, prepare a campaign draft and flag a pricing change. It shouldn’t publish a social post, alter a CRM record, move paid budget or promise a customer something without a named approver.

“By giving Dots their own identities, they become an extension of our team.”

Holly Li, OpenAI product team member, at DevDay, as reported by Barron’s.

Extensions of a team need the same things teams need: a defined job, access only to the systems needed for that job, a record of what changed, and a person who owns the call when a customer-facing decision is involved.

Start with one recurring workflow

Start with recurring work that has a visible input and a clear review point.

A small team could ask a Dot to watch customer interviews, pull recurring objections into a weekly brief, compare those objections with the product page and prepare proposed updates for a product marketer to review. Nothing ships on its own. The team measures whether the brief arrived sooner, used better evidence and removed a handoff.

The value sits in whether the work stays connected from signal to decision.

There is no performance evidence yet. OpenAI hasn’t published results on campaign output, time saved, accuracy or customer impact, and it hasn’t released a standalone Dots product document alongside the keynote. The live demo also had glitches, Axios noted.

Dots is an early interface for a promising workflow category. Customer-facing work still needs a human owner.

The immediate job is simpler. List the work that already moves through research, content, operations and approval. Pick one lane where an agent can carry context without carrying final authority. The teams that define that boundary early will have a better system to scale when persistent agents become normal.