Asking AI to rephrase a text and authorising software to modify a document are two different uses of artificial intelligence. An announcement made by Apple on 27 October 2026 brings these possibilities closer together in Pages, Numbers and Keynote. To understand what is changing, it is important to distinguish between the assistant built into the app and connecting an external agent.
Apple plans to update Creator Studio in November, without giving a specific date in its announcement. The assistants announced in beta use Apple and OpenAI models: writing in Pages, working on slides in Keynote, and calculations and charts in Numbers. They are designed to take the open document into account.
Access requires a device compatible with Apple Intelligence and at least iOS 27, iPadOS 27 or macOS 27; the beta is limited to US English. MCP support, meanwhile, requires macOS 27. Free versions of the apps remain available: the new features described are for subscribers. Apple’s announcement and terms.
An assistant in the document, an agent connected to the app
The built-in assistant offers the most direct approach: someone works in their document and makes a request there. Opening the app up to an external agent makes another way of working possible. Apple says these agents will be able to read content, make changes and carry out actions in all three apps.
MCP stands for Model Context Protocol. Its documentation describes an architecture in which an AI application connects to programmes that give it access to information or tools. The protocol organises this exchange; on its own, it does not determine how a model reasons or how an application carries out its work.
This is a practical distinction. A connector capable of reading a spreadsheet is not automatically a reliable analyst of its figures. And a tool capable of saving a change does not guarantee that the change meets the need. The capabilities made available, the agent’s behaviour and the checks on the result all need to be considered together. MCP’s official architecture.
What automation could spare you from
Take a hypothetical example: an association has an attendance spreadsheet and is preparing a presentation for its general meeting. In a wholly manual process, someone selects the data, calculates the changes, chooses a chart and then transfers the results to the slides. A connection between tools could reduce some of these transfers.
The interesting benefit, then, is not just producing sentences. It is the possibility of retaining the link between information and where it is used. But this example is not a demonstration of Creator Studio: we have not tested this process, and the announcement proves neither its accuracy nor the time saved in this specific case.
To assess such a use, one useful check would be to compare the original figures with those in the presentation, verify the periods selected, and then examine the chart titles. A change can be calculated correctly yet still be explained poorly if the text omits a change in scope.
Reading and editing require different checks
The MCP tools specification recommends making tools visible to the user, signalling when they are being used and allowing human intervention. It also recommends validating inputs, controlling access and checking results. These principles describe a framework; they do not prove exactly how Apple has implemented it in its apps.
So before entrusting a work document to an agent, the practical questions remain the same: what data can it access? What changes can it save? How can you inspect what has been changed? And which services process the request? The answers should be found in the product documentation and the settings actually offered, rather than inferred from the acronym MCP alone. Tools specification and security recommendations.
An announcement to assess through verifiable tasks
For a French-speaking user, the announced language restriction is an essential condition, not a footnote. For a team, the question is also whether it wants occasional help with writing or a series of actions carried out on files.
A good trial will start with a document whose data and expected outcome are already known. It will make it possible to compare the work produced, the corrections needed and the time actually saved. The promise of an app that is easier to control becomes interesting when it stands up to this check; the announcement alone does not yet make it possible to put a figure on it.
Illustrative photo: an Apple computer; the image does not show Creator Studio’s new features.
