Siri AI Opens to All Developers: Is Your App Ready for AI Assistants?

On 3 November Apple opened Siri AI to every app, though App Store releases wait until spring. How App Intents, entities and app skills work, and what product teams can do now.

Published 6 min read

On 3 November 2026, Apple opened Siri AI to every app developer. A Developer Preview in the iOS, iPadOS, macOS and visionOS 27.3 betas lets any app hand its own actions and content to Siri. Until now, apps could only plug into a smaller set of predefined schemas. According to Macworld, Apple's post put it plainly: the preview "allows any app to bring its content and actions directly into Siri AI by using App Intents".

The catch is timing. Developers can build and test today, but apps using the preview cannot ship to the App Store until spring.

This matters beyond Apple. It is one more sign that software is getting a second user: an AI assistant that reads what your product can do and acts on the user's behalf. This article covers what Apple opened, how it works, what is still missing, and what product teams can prepare now, whatever platform they build on.

What Apple opened on 3 November

Before this preview, Siri could work with third-party apps mainly through fixed categories of actions. 9to5Mac reports that the new initiative lets any app create Siri AI integrations, including "app skills" that teach Siri to coordinate several steps.

The examples Apple and the press use are ordinary on purpose: checking a delivery, booking a table, placing an order, or changing a ride's pickup address during a Live Activity without naming the app. eWeek notes that the preview also lets Siri find app content that does not fit an existing schema, through expanded Spotlight indexing.

Apple's own apps get the same capabilities. PPC Land, which went through Apple's documentation in detail, lists Calendar, Files, Mail, Maps, Messages, Notes, Photos, Reminders, Safari and others.

How it works: actions, content and skills

The integration is built on App Intents, the framework Apple already uses to connect apps to Siri, Spotlight and Shortcuts. It has three parts.

Intents: what the app can do

An intent is one action, such as "track flight" or "add item to cart". According to PPC Land, the preview adds a Swift macro, @AppIntent(.tool), that turns an intent into a tool Siri can call, with no new entitlement or separate program needed. Exposure is opt-in for each action: Siri can only use what the app explicitly makes available.

Entities: what the app knows about

Entities are the objects an app works with, such as a flight, an order or a note. Each property gets a description so the model understands it. Developers can require authentication before an entity is shown, or mark a property as internal so Siri never sees it. Entities go into the on-device Spotlight index, so Siri can find them.

Skills: how to combine actions

A skill is a short Markdown file, SKILL.md, that explains to Siri how to chain an app's intents for a larger task. PPC Land reports a name limit of 64 characters, a description of up to 1,024 characters and a body of up to 10,000, checked by Xcode at build time. The format is described as a subset of the open Agent Skills format.

One detail deserves attention: skills are advice, not programs. Siri does not necessarily run the listed steps in order, and results can vary between runs. Apple's new Siri Evaluations framework lets developers script requests and measure the answers, and PPC Land notes that each set of scenarios should be run more than once. Xcode also gains a Siri Integration Advisor that comments on an implementation.

What you cannot do yet

The preview is for building and testing only. Several limits matter when planning:

Question

Current answer

Can users get it from the App Store?

No. Submissions are planned for spring, with no exact date.

Who can test?

Development builds on a supported device, and internal TestFlight testers. External testers get nothing.

Terms and usage policies?

Not published yet; Apple says details will follow.

Languages?

English, French, Japanese, Korean, Portuguese and Spanish, with more announced as "soon". Persian and Arabic are not on the list.

European Union?

Siri AI is unavailable on iOS, iPadOS and watchOS in the EU; Mac and Vision Pro are not affected.

Usage limits?

Server-side features have daily limits; Apple says expanded access will cost a fee later.

The EU gap goes back to June. In its newsroom statement, Apple said that because of the Digital Markets Act, developers in the EU will not be able to test the new Siri AI features on iOS 27, and that there is "no timeline" for availability there. Apple says Siri AI is private by design, using on-device processing and Private Cloud Compute.

The bigger shift: your app as a tool

Strip away the Apple branding and the pattern is familiar. Assistants and AI agents increasingly do not click through your screens. They read a list of named actions and described data, decide which ones fit the user's request, and call them. Apple's choice of a format close to Agent Skills shows how quickly these conventions are converging across vendors.

That changes what "a good app" means in three ways:

  1. Descriptions become part of the product. If a model chooses actions by reading their names and descriptions, an unclear description is a bug. "Update" tells the assistant nothing; "Change the delivery address of an open order" tells it exactly when to use the action.

  2. Exposure becomes a decision. Opt-in per action means someone has to decide, deliberately, which actions an assistant may run and which data it may see. That is a product and security choice, not only an engineering task.

  3. Testing becomes statistical. When the same request can produce different paths, one passing test proves little. Teams need sets of realistic requests, run several times, with a threshold for "good enough".

What product teams can do now

You do not need an iOS 27.3 beta to start. These steps help with Siri, and with any other assistant that may call your product later:

  • List your top ten actions. Choose the tasks users repeat most, and write each one as a clear verb plus object.

  • Describe your data. For each main object, write one plain sentence about what it is and which fields matter.

  • Sort actions by risk. Reading and searching are low risk. Sending, paying, deleting and changing other people's data should need a confirmation from the user. We explained a simple way to sort them in When Should an AI Agent Ask Before It Acts?.

  • Hide what should stay hidden. Decide which fields an assistant must never see, and require sign-in where data is personal.

  • Collect test requests. Write 30–50 real requests in your users' own words, including vague and messy ones, and keep them for evaluation.

This is also the approach we follow in Hamroshd: its AI assistant works inside the team's workspace and asks before it changes anything. Whatever tools you use, an assistant that can act needs clear actions, clear limits and a person who approves the risky steps.

Sources