That conflict is more important than the marketing language around “vibe coding.” The app-builder skill can create and alter Dataverse tables, forms, views, charts, sitemap entries, JavaScript validation rules, security roles, business process flows, and generative pages. A tool that can write that breadth of application metadata is no longer merely an assistant for drafting screens; it is an automation path into a production business-app estate.
Microsoft published the monthly roundup on September 17, alongside updates to canvas-app templates and controls, the opt-in refreshed interface for model-driven apps, generative-page data connectivity, and server-side flow search. The practical split is clear: several changes are safe productivity improvements, while the AI authoring features require the same review, test, and change-control practices that administrators apply to any other automated deployment tool.
The model app-builder’s status is not settled by the announcement
The September roundup calls the model app-builder skill generally available and says that makers can use GitHub Copilot CLI, Claude Code, or a supported AI coding tool to create or edit a whole model-driven app from natural-language requirements. Microsoft says the process produces a plan for review before it writes artifacts, and that its output remains ordinary Power Apps and Dataverse components that can be modified later in the standard designers.
But Microsoft Learn’s dedicated documentation, last updated August 28, still labels the identical model app-builder skill as preview. It explicitly says the capability is under active development, warns that its app-spec schema and CLI options can change between releases, and says it does not yet support every model-driven-app artifact or concept. Microsoft’s own September roundup also links to the learning module under the title “Build and edit model-driven apps with AI code generation tools (preview),” even while its feature headline says generally available.
The reasonable conclusion is not that the feature failed to launch. Microsoft’s September announcement establishes that its product team considers the app-builder ready for wider use. The issue is that the supporting record has not caught up: customers lack a consistent answer on support expectations, change stability, and the boundary between the production-grade workflow and the still-preview components it uses.
Administrators should therefore classify the model app-builder as a controlled pilot until Microsoft updates the Learn documentation and resolves the preview designation. Keep it pointed at developer or sandbox environments, require a human review of the generated specification and dry-run plan, and deploy through established solution-management pipelines rather than allowing direct edits in shared production environments. The plan approval step is valuable, but it is not a substitute for reviewing security roles, table relationships, JavaScript web resources, and business rules created from a broad-language prompt.
Microsoft Learn also notes a more specific limitation: manual changes made after generation can affect later AI-assisted edits. That makes source control, solution exports, and a documented baseline essential before asking the agent to revise an existing app. The most dangerous failure here is not an obviously broken generated app; it is a later agent-led update that quietly overwrites a maker’s post-generation customization because that change was not reflected in the next app specification.
Generative pages move beyond Dataverse, but the connector path remains preview
The more concrete expansion in the update is that Power Apps generative pages can now consume data through Power Platform connectors rather than being confined to Dataverse tables. Microsoft says makers can bind a page to an existing connection reference for sources including SharePoint and SQL, either through the generative-page designer or through the external AI coding workflow.
For organizations whose working data is distributed across Microsoft 365, SQL Server, SaaS applications, and line-of-business systems, this removes a common reason to duplicate data into Dataverse just to produce a tailored app experience. It can allow a model-driven app record to retain its governed Dataverse core while displaying connected operational data beside it.
There is a clear caveat: Microsoft labels connector support for generative pages as a public preview, and the corresponding Learn documentation says preview features are not intended for production use. It also says each connector-backed page requires an existing Power Platform connection, and that teams should test the generated page using the same identities and permissions their users will have.
That identity requirement deserves attention. A page that works for its maker can still fail, expose unexpectedly broad information, or produce incomplete results when opened by a regular user. Before testing a connector-backed generative page, admins should inspect its connection reference, data-loss-prevention policy treatment, sharing configuration, connector-specific permissions, and whether the intended end-user identity can access the underlying source. The agent’s ability to propose and deploy a binding does not alter any of those controls.
Microsoft also now allows a generative page to appear inside a section or tab of a model-driven-app form. The active record’s recordId can be passed to that embedded page at load time, making it possible to combine standard structured fields with a custom AI-generated view. This is more useful than adding a separate navigation page, particularly for case-management, customer-service, or operations apps where users need both governed record fields and a contextual dashboard in one place.
Canvas app modernization comes with formula migration work
For canvas-app makers, Microsoft has added six Fluent 2 screen templates to the newer screen-creation experience, joining three templates released earlier. The templates replace older out-of-box designs based on classic controls and give makers responsive layouts and editable modern controls as starting points.
The higher-impact change is bulk updating of eligible modern controls from the Power Apps Studio control-update surface. Microsoft says “Update all” works on the current screen, can enable required modern-control settings, and preserves undo and redo. That will save time in apps with many controls, but it should not be treated as a cosmetic refresh.
Microsoft Learn documents changes introduced to modern controls since February 2026 that include renamed properties, enum-based values, altered defaults, and changed event behavior. Examples include FontColor becoming Color, FontSize becoming Size, and string settings being replaced by typed values such as FontWeight.Bold or Align.Left. A formula that previously used a plain text value may need an explicit enum reference after the upgrade.
The behavioral changes can be more consequential than the property renames. Microsoft says a modern Text Input control’s OnChange now fires on focus loss rather than every keystroke, while a Combo Box fires on every selection or deselection. A canvas app that runs validation, filtering, calculations, or side effects from those events can appear to migrate successfully yet behave differently for users.
The defensible procedure is to update a copy of the app, screen by screen, then test every workflow involving updated controls—not simply open the app and confirm that it renders. Makers should pay particular attention to OnChange, TriggerOutput, default values, read-only states, formulas with quoted enumeration values, and any live-search behavior. Microsoft recommends directly referencing a text control’s output property when a formula needs per-keystroke updating, rather than relying on OnChange.
Microsoft also added a modern progress bar, avatar, and spinner, while reporting fixes in Date Picker behavior and formula preservation for Checkbox, Rating, and Toggle controls. These are welcome incremental improvements, but the migration work is the operational story: bulk updating controls increases the need for deliberate regression testing, rather than eliminating it.
Model-driven UI changes are opt-in, and density depends on the refresh
Model-driven apps receive the header-and-navigation refresh in version 2609.1, according to the Power Platform Blog. The revised layout simplifies the header and sitemap, lets the form header scroll with the page, and leaves the command bar as the fixed top element. A condensed sticky header appears beneath that command bar after the primary form header scrolls out of view.
Microsoft has kept this refresh opt-in at general availability, which is the right choice for organizations with established training materials, tightly controlled browser automation, or user groups sensitive to navigation changes. Apps that have already enabled the feature receive the latest refinements automatically, so app owners should verify the refreshed layout in their actual forms and custom command scenarios rather than assuming the GA label means no change is pending.
The new display-density setting is a public preview and depends on enabling the header-and-navigation refresh first. It offers Comfortable, Cozy, and Compact choices; admins can establish a default or turn the setting off, while individual users can choose their own density through personal settings. This is a usability feature rather than a data or platform change, but it can affect the visible portion of grids, forms, and dashboards enough to expose cramped custom layouts.
Organizations should roll out the header refresh first to a representative app and user group, then evaluate Compact density with the busiest forms and widest grids. If a business process relies on dense record inspection, Compact may be welcome. If teams use heavily customized forms designed around a particular viewport, it may expose alignment and readability issues that the default Comfortable option masks.
Power Automate gets a quieter but practical improvement
Power Automate’s addition of server-side search in My Flows may be the least flashy update and the one most immediately useful to flow owners. Microsoft says search now queries the service’s full set of flows in the current environment rather than filtering only the subset already loaded in the browser.
For administrators, makers, and support teams responsible for environments with hundreds or thousands of flows, that changes a frustrating discovery task into a more dependable one. A missing result previously could mean a flow was absent, hidden by ownership or permissions, or simply never loaded into the browser’s local list. Server-side search does not solve access-control constraints, but it should remove the last of those explanations.
Microsoft also added quick-start cards to the Power Automate home page. They reduce the number of clicks between an automation idea and the flow-creation experience, but they do not replace governance. Organizations that use environment routing, managed solutions, connector controls, and data-loss-prevention policies should ensure those guardrails are evident before inexperienced makers turn a home-page prompt into a production workflow.
The September update gives Power Platform teams real gains in app authoring, interface flexibility, and flow administration. Yet the document trail around the model app-builder skill means the immediate action is not to hand it unrestricted production access: test it in a governed environment, preserve a recoverable solution baseline, and wait for Microsoft to make its “generally available” announcement match its own technical documentation.