Two OpenAI announcements point in the same direction: ChatGPT is no longer to be primarily a dialogue window for individual users, but a platform on which teams work together and on which software runs permanently in the background.At DevDay, ChatGPT was opened up into a team platform.In parallel, agents called Dots were introduced that remain permanently active.
Shared workspaces instead of individual chats
The central element of the platform opening is shared workspaces, called "Space" at OpenAI.These are complemented by collaborative documents and slides, i.e. presentations that several people can edit together.The difference from the previous model is less technical than organisational: results no longer arise only in a person's private chat history, but in a place where a team can find them again.
This is supplemented by an open plugin system.It builds on MCP, a protocol through which language models access external tools and data sources in a standardised way.What is new here is automation via MCP events: workflows can therefore be triggered by events and not only by manual input.This shifts the role of the model from an answer machine to a trigger of processes.
Integration into existing communication
Both announcements name the same entry points: Slack and Microsoft Teams.That is noteworthy because it means teams are not expected to change their working environment.Instead, the functions move to where communication already takes place.For introduction in the company, this significantly lowers the barrier, but at the same time shifts the governance question into channels in which work has so far tended to be more informal.
On the commercial side, an enterprise marketplace with 32 partners is mentioned as well as a new Pro-500 plan.Details on terms or availability are not evident from the sources, which is why restraint is appropriate here.All that is recognisable is the intention to build an ecosystem of third-party providers instead of delivering every integration itself.
Dots: agents that do not wait for a question
The second building block goes a step further.According to OpenAI, Dots are permanently active agents that work on their own cloud computers.Examples given are fixing bugs or sending forgotten invoices.The decisive addition is that this partly happens before anyone asks for it.
Behind this, technically, is a simple but consequential pattern: when inactive, the agents search in the background with read permissions for possible approaches to work.The restriction to read permissions in this phase is the most interesting detail in the announcement, because it separates observation from execution.The agents are in turn addressed via ChatGPT, Slack and Teams.Dots is positioned against Meta's Muse, which shows that permanent agents are evidently not remaining a one-off approach.
How the two announcements fit together
The sources report independently but describe two halves of the same development.The platform opening provides the framework: shared workspaces, standardised connection via MCP, event-based triggers and presence in the communication tools.The permanent agents provide the actor that uses this framework.Without event automation and tool access, a permanently running agent would be largely ineffective.
The two announcements do not contain any divergent details.They differ in level of detail: the platform announcement is concrete on functions and plans, the agent announcement on the way of working.What is missing in both cases are statements on detailed permission management, logging or data storage.These very points decide in practice whether such a setup is viable in a medium-sized company.
Obvious use cases around the website
If one transfers the described patterns to everyday work, fields emerge in which event-driven automation already makes sense today:
- Editorial: Notices about outdated content, missing alternative texts or inconsistent metadata, triggered by changes in the content management system.
- Support: Pre-structuring incoming enquiries together with references to existing documentation, triggered by a new ticket.
- Operations: Preparation of error messages and log anomalies with a suggestion for classification, before someone looks manually.
- Administration: Reminders of open transactions, as in the example of the forgotten invoices.
The common denominator: these are tasks with a clear trigger, verifiable results and limited potential for damage in the event of errors.
What should be clarified in advance
Anyone who wants to use permanent agents needs a robust answer to three questions.First: which systems may an agent read and which may it change?The separation mentioned in the sources between reading in the background and active action is a sensible model.Second: where is it documented what happened automatically?Third: who checks results before they have an external effect, for example on the website or towards customers?
For web and software projects, the focus thus shifts from the question of which model is used to the question of what clean interfaces and clear permissions look like.Systems such as TYPO3 benefit from this when content, workflows and status information are accessible via defined APIs instead of only via the interface.
Anyone who establishes these foundations now can gradually add automation without tying their own architecture to a single provider.