ChatGPT Dots Keep Working After the Chat Ends

ChatGPT Dots can keep working after the conversation ends. For ecommerce teams, that puts a familiar problem in focus: a customer may have an answer while the work behind it remains unfinished.

A delayed parcel still needs checking. Repeated sizing complaints still need investigating. A supplier still owes someone an explanation. These loose ends are a useful way to assess the always-on agents featured in OpenAI's update for September 28-October 2.

What OpenAI announced

Dots take on ongoing responsibilities in ChatGPT. According to OpenAI, a dot uses connected apps, has its own cloud computer and browser, and brings back results or decisions that need attention. Availability is rolling out gradually to eligible Pro, Business Premium, and Enterprise accounts; merchants should check OpenAI's current access requirements.

The practical shift is continuity. Instead of asking for a fresh answer each time, a user can give an agent a defined responsibility to carry forward between conversations. For support operations, that invites a useful question: which unfinished tasks deserve that kind of follow-through?

Three tasks worth examining

The following are possible merchant workflows, not announced Dots features or ready-made Shopify integrations. Each would need suitable connected tools, access to the relevant records, and clear permission to use them.

Find the issue behind repeated questions. Imagine several shoppers asking whether a jacket runs small. A useful assignment would be to review the relevant conversation records, group the recurring questions, and prepare a brief for the merchandising team. The output could point to a size guide that needs attention, with links back to the evidence.

That brief should distinguish a handful of comments from a demonstrated product problem. It should also separate what customers said from the agent's interpretation. Someone responsible for the product page can then decide whether the evidence warrants a change.

Keep an unresolved delivery task visible. Suppose a customer has received the latest verified order status, but a team member is still waiting for an internal shipping update. An ongoing task could check the authorized source again and surface the new information when it arrives.

The assignment needs a stopping point: report the update, flag that nothing has changed by an agreed time, or request help if access fails. It should not turn silence into an invented delivery date, refund promise, or replacement decision. Those actions require the appropriate information and authority.

Prepare the next shift's handover. Another possible use is a short brief on unresolved cases: what the customer asked, what has already been checked, and what decision remains. The value would lie in preserving the thread of the work so the next person can act without reconstructing it.

A useful handover would link to the underlying records and state what is missing. An uncertain conclusion should remain uncertain. A polished summary is only helpful if the next operator can verify it.

Give the agent a clear assignment

These examples depend on access and instructions, not simply on an agent staying online. OpenAI's supplied documentation does not establish a ready-made Dots connection to a merchant's Shopify store or customer inbox. A proposed workflow must be checked against the tools actually available.

The same care applies to action permissions. OpenAI describes controls that review actions against instructions and permissions. Permission to research an issue or draft a message should not be treated as permission to send it, change an order, or issue a refund.

Start with a bounded assignment: the records to examine, the question to answer, the output expected, and the decisions reserved for a person. Set a clear point for reporting a blocked task. Otherwise, an agent can produce activity without producing an answer the team can use.

Where Kolton fits

The connection between these ongoing agents and ecommerce support is the customer conversation that creates or resolves the work. Kolton grounds sales and support answers in connected Shopify catalog, stock, and order information across web chat, email, WhatsApp, and Instagram/Facebook. Its actions follow the merchant's configured permissions, with help requested when information or approval is needed.

An ongoing agent and a customer conversation share a practical requirement: usable context. A follow-up brief needs the right records, just as an order-status answer needs current order information. This is an operational connection, not an announced integration between Dots and Kolton.

Start with one unfinished job

A merchant evaluating Dots could begin with one recurring task that currently gets lost between replies. Define what a useful result looks like, keep consequential decisions within the agreed permissions, and review whether the output helps the team finish the job.

Measure the quality of the follow-through: were the records accurate, were open questions preserved, and did the result make the next action clearer? That is a more useful test than how much text the agent produced.

For the conversations themselves, see how Kolton uses connected Shopify context for sales and support.