What the roadmap promises
Microsoft’s roadmap entry 576284 lists the feature as In development, with preview targeted for October 2026 and general availability for December 2026. Its metadata identifies Gen page v307, the web platform, model-driven apps, and admins or makers. It also lists Worldwide Standard Multi-Tenant, GCC, and GCC High cloud instances. Those are planning details—not confirmation that every listed environment already has the feature. Microsoft explicitly describes roadmap dates and descriptions as estimates subject to change.
For now, the important distinction is documented preview support, not a general-availability launch. Microsoft warns that the preview is not intended for production use and may have restricted functionality.
An operation, not just another table edit
Microsoft Learn distinguishes an explicitly invoked custom API from a plug-in that automatically runs when a Dataverse row is created, updated, or deleted. Those event-triggered plug-ins already execute when a generative page performs the corresponding operation; makers do not need to attach them separately. A custom API is for invoking a named operation at a chosen moment.
That distinction matters architecturally. An approval button can call an existing business operation rather than treating approval as merely changing a field. The generated interface provides the interaction; the prepared API supplies the server-side implementation. That is an interpretation of Microsoft’s documented workflow, not a claim that AI has validated the underlying business rules.
How makers connect the page
Microsoft’s browser-based procedure is straightforward:
- Open a model-driven app for editing.
- Select Add page > Generative page, then Describe a page.
- Choose Add > Custom API (preview) and select the prepared API.
- Describe the trigger, input values, and response handling.
- Select Generate page and review the result.
The API must already exist. Microsoft’s prerequisites describe creating a Dataverse plug-in, registering a custom API with the Plug-in Registration Tool, and associating it with the plug-in type.
There are compatibility boundaries: global and table-bound actions and functions are supported, but APIs using EntityCollection binding or Entity parameters are not. Generative pages remain limited to model-driven apps, rather than canvas apps.
The code-first route—and the review that still matters
Microsoft also documents the capability through AI code-generation tools. During planning, the agent lists supported APIs in the selected environment, asks the maker to choose one, incorporates its request and response properties, and includes the API binding in deployment. Microsoft instructs makers to review the API inputs before approving the build and test using their users’ identities and permissions.
Availability differs between authoring routes. Microsoft documents code-first authoring worldwide in public clouds, while browser authoring is limited to environments in the United States, Great Britain, Australia, and Singapore.
Security remains a testing responsibility, not a checkbox supplied by generation. Custom APIs execute in the signed-in user’s context and remain subject to Dataverse security. A sensible acceptance test therefore checks both the intended result and how the page behaves when the user lacks permission—not merely whether an attractive Approve button appears.