Published November 19, 2026
Apple Opens Siri AI to Every App: What App Intents Mean for a Flutter Codebase
On November 3, Apple opened the Siri AI developer preview to every app. The change landed in the iOS 27.3, iPadOS 27.3, macOS 27.3, and visionOS 27.3 betas, and the short version is that any developer can now teach Siri actions that execute inside their own application instead of handing the request off and hoping the user goes looking for the icon.
The shift is narrower than "Siri controls everything" and considerably larger than it first sounds. Until the preview, Siri's third-party reach was a fixed list of tasks Apple defined: send a message, get directions, start playback. Anything outside that list ended at the app icon and a tap.
The mechanism is App Intents, the same framework that already powers Shortcuts and Spotlight. An app declares what it can do, gives each action typed parameters, and supplies the phrases Siri should listen for. A customer asks in plain language, Siri resolves the intent, your code runs, and the answer comes back with your branding rather than a generic system response.
Very Good Ventures published a working demonstration on the Scooter's Coffee app. A customer asks Siri for the menu, narrows it to lattes, and Siri pulls the answer from the app without the customer ever launching it. That is the shape of the feature: the assistant becomes a surface your product appears on.
What this looks like inside a Flutter project
The first fact that matters is that a Flutter app on iOS is an iOS app. App Intents are a Swift surface, and nothing in the framework cares that the pixels are drawn by Impeller. The integration pattern is a thin native layer, not a second application.
On the Swift side you declare an AppIntent with typed parameter properties, a shortcut provider that maps spoken phrases to that intent, and a perform method that returns a result the assistant can speak or render. Most of the design work is choosing which capabilities deserve an intent at all: reorder the usual, check the delivery, move the appointment, book the slot.
That layer then calls into Dart. A method channel or a Pigeon-generated interface hands the request to repository code you already have, which is where authentication, caching, retries, and error handling live. When Dart answers, the Swift intent wraps the result in a dialog or a snippet. Nothing here requires rewriting the app in Swift, and nothing here breaks hot reload for the rest of the UI.
The part teams underestimate is vocabulary. Siri matches an utterance against an intent and its parameters. If the parameter is named item and customers say "the usual," you have work to do: entities with display representations, synonyms, and disambiguation prompts so the assistant can ask a sensible follow-up question instead of failing.
The window is already running
The preview is deliberately restricted. Integrations run only in development-signed builds and for internal TestFlight testers, never for external testers or App Store users, and only on devices that support Apple Intelligence. Apple has also said the App Intents API may still change, which is worth planning around rather than complaining about.
- June 8, 2026: Apple unveils the new Siri at WWDC.
- September 14, 2026: Siri AI reaches users in beta.
- November 3, 2026: the developer preview opens to every app.
- December 2026: iOS 27.3 ships, but app-level Siri features stay dark.
- Spring 2027: App Store submissions open and customers get access.
Read the last two entries together. The operating system arrives in December, the store opens months later. That gap is the build window, and it is the same length for every competitor. Teams that treat the preview as a curiosity will be starting integration work at the moment their rivals are submitting.
There is a measurement problem attached to the feature. Requests that begin with Siri arrive without the referral trail a normal session carries, so a reorder can look organic or can look like nothing at all. Decide before launch what an intent-originated conversion means in your analytics, and give each intent its own signal so the report is not guesswork.
There is a product question too. If a customer can reorder without opening the app, which screens become optional? The honest answer for most teams is that an intent should map to a task rather than to a screen. That framing keeps the native layer small and stops the project from turning into a second, worse front end maintained in Swift.
None of this is exotic engineering. It is one Swift file, one channel, and a decision about which five tasks matter enough to expose. The genuinely expensive part is starting in April, when the only thing left to do is wait for a review queue.