Apple on November 3, 2026, gave every developer a way to plug an app into Siri AI, releasing a Developer Preview with the iOS 27.3, iPadOS 27.3, macOS 27.3 and visionOS 27.3 betas that lets the assistant carry out an app's own custom actions, from checking a delivery to placing an order. Until then, Siri AI could act inside third-party apps only where a task fitted one of the categories Apple had defined in advance. The change will not reach customers for months. App Store submission opens in the spring, and Apple has yet to publish the terms and usage policies that will govern what developers build.

In Short

Siri AI, Apple's new assistant, could already do a few things inside other companies' apps, such as sending a WhatsApp message, but only for everyday jobs Apple had listed in advance. Now any app maker can teach it the app's own tasks, like checking an order or booking a table, so you can ask in your own words and Siri AI does the job inside that app. Developers can test this in Apple's beta software from November 3, but apps on the App Store cannot offer it until next spring, and Apple has not yet said what rules will apply.

What Apple published on November 3

The announcement went up on Apple's developer news site at 10:00 a.m. Pacific time on November 3, two hours after a separate post said the 27.3 betas of all six Apple operating systems, together with Xcode 27.3, were available. "Starting today with the iOS 27.3, iPadOS 27.3, macOS 27.3, and visionOS 27.3 betas, a new Developer Preview enables any developer to integrate their app with Siri AI, giving businesses and developers new ways to reach and serve their customers," the post opens.

Apple pitched the change as distribution. Siri AI, the post says, is "designed to be deeply integrated into Apple's operating systems and the apps people use every day, including the apps from our vast ecosystem of developers." It adds that "thousands of developers have already built integrations with Siri to let users ask for directions, manage a to-do list, send a message, or catch up on an audiobook." Then comes the commercial argument: "People around the world are turning to Siri AI to get things done every day, which means more customers can discover and use apps just by asking." No figure accompanies the claim. Apple did not say how many people use Siri AI or how many of their requests reach third-party apps.

Three changes sit at the center of the preview. The first is custom actions. "Today's preview allows any app to bring its content and actions directly into Siri AI by using App Intents, making it easier for all apps to tap into Siri AI's newest capabilities, including personal context understanding, onscreen awareness, natural language, and more," according to Apple. The second is a new layer of instructions, which Apple calls skills. The third concerns content: Siri AI can now find anything an app has placed in the device's Spotlight index, not only content that matches one of Apple's templates.

Apple's own software went first. The company said it had started rolling out the capabilities across 15 of its apps - Calendar, Files, Finder, Journal, Mail, Maps, Messages, Music, Notes, Photos, Preview, Reminders, Safari, Shortcuts and Weather - "so users can do things like find a coffee shop they visited last week or archive files that haven't been opened in over a year." Those actions double as reference material for developers.

The preview does not cover every platform. Siri AI itself runs in beta on watchOS 27.3, and a tvOS 27.3 beta shipped the same day, but Apple's documentation limits the Developer Preview to iOS, iPadOS, macOS and visionOS, on devices that support Apple Intelligence. During the preview, the App Intents API and Siri's behavior may change, the documentation warns.

From 13 schema domains to any custom action

App Intents is the framework Apple uses to let apps describe their actions, called intents, and their content, called entities, to system features such as Siri, Spotlight and the Shortcuts app. Until this week, Siri AI could drive a third-party app only through app schemas: predefined templates for common tasks. Apple's documentation lists 13 primary schema domains that work with Siri and Apple Intelligence - audio, calendar, camera, clock, mail, maps, messages, notes, phone, photos, reminders, system and in-app search, and video. Nine more, covering books, browsers, files, journaling, presentations, document readers, spreadsheets, whiteboards and word processors, work only in Shortcuts. A messaging app that implemented the send-message schema could be operated by voice. A delivery app's order tracker could not, because no schema describes it.

That boundary was visible when the assistant shipped. At the September 14 launch of Siri AI, the third-party actions working on day one were texting through WhatsApp and playing a title in Audible, while drafting mail in Microsoft Outlook, searching an assignment in Notability and adding a restaurant to Tripsy were listed as coming. Messaging, audio and mail all have schemas.

The preview removes the template requirement. A developer marks an intent with a Swift macro, @AppIntent(.tool), which makes it conform to a new ToolIntent protocol and exposes it to Siri as a tool the assistant can call. The vocabulary is borrowed from the wider AI industry, where a language model is handed a list of tools, each with a name, a description and parameters, and decides which to invoke. Intents that already conform to a Siri-supported schema stay as they are. Those built on a Shortcuts-only schema, such as creating a journal entry, can be promoted with @AppIntent(.tool, schema:). No new entitlement and no separate developer program are needed, according to Apple's documentation.

Exposure remains opt-in, action by action. Siri can only find content and perform actions that an app makes explicitly available, the documentation says, and which intents to hand over is left to the developer.

Describing an app to a model

Because a model chooses between tools, the words attached to each one carry weight. Siri relies on the title and description of an intent, and on a title and description for every parameter, to understand what an action does and to tell it apart from similar actions, according to the documentation. Apple's sample is a shopping app's add-to-cart intent, with an item and a quantity that defaults to one.

Entities - an app's content objects, such as a product, a flight or a bank account - get the same treatment, with a description on each property. Developers can require authentication before Siri presents an entity or a single property, and can mark a property as internal, which hides it from Siri, Shortcuts and other system features altogether. Apple's bank account example lets Siri acknowledge that an account exists, asks for authentication before revealing its nickname, requires authentication on the device itself before showing the balance, and keeps the routing number inside the app.

A new generable trait lets Siri build an entity directly from what a person says, without asking the app to look it up first. The documentation's example is a flight search assembled from a request for an economy flight from San Francisco to Los Angeles next Friday, with origin, destination, date and cabin class taken from the sentence.

Finding content runs through Spotlight. To let Siri locate an app's material, developers add their entities to the on-device Spotlight index, with an entity query as a fallback for apps that do not. That index is the semantic layer the system orchestrator Apple described at WWDC on June 8 draws on, alongside an app toolbox that runs on device. "And Siri AI can now surface any entity in the Spotlight semantic index, not just the ones that fit an existing schema," the post says.

Three further hooks shape what Siri understands. Onscreen context, which links an app's views to its entities so a person can refer to an onscreen photo conversationally, now works for custom entities as well as schema-based ones. Donations, in which an app reports actions a person takes, give Apple Intelligence behavioral cues to spot routines and resolve vague requests; Apple's overview describes someone who checks the weather in an app every morning, after which the system can suggest the same action proactively. And cross-app requests depend on entities offering a transferable representation. Apple's example is a person asking Siri to email a flight from Seattle to San Jose to a partner, which carries data from an airline app, through Contacts, into Mail.

When the app has to be named

The documentation spells out when Siri can act without being told which app to use. A person does not need to name the app while it is in the foreground, while one of its Live Activities is running, or when continuing a conversation in the Siri app that already mentions it. Apple illustrates this with a ride-share trip: a person books a ride, follows it in the app's Live Activity, then asks Siri to move the pickup to a new address without naming the service. When the app sits in the background, the person has to mention it by name. Apple's example is a question about whether sneakers the person liked in a named shopping app are on sale.

Requests can also cross apps. In another example, Siri finds a pair of favorite sneakers in a shopping app, identifies the person's partner in Contacts and sends the details through a named messaging app. Results come back in card snippets that developers build with their own views, which carry the app's branding, according to the documentation.

Skills, written in Markdown

"For complex requests, we're also introducing skills to let you teach Siri AI how to combine several actions into a single experience," Apple says. A tool intent performs one action. A skill is a page of natural-language guidance telling Siri how several tools fit together, which one to prefer when more than one could apply, what to avoid, and what domain knowledge the app requires. The documentation's example is a shopping skill that covers searching for items, adding them to a cart, choosing payment and delivery methods, reviewing the cart and confirming the purchase.

The format is strict about its wrapper and loose about its contents. Each app target holds one skills bundle, with one folder per skill and a single SKILL.md file inside each. The file opens with two required fields: a name of 1 to 64 lowercase letters, digits and hyphens that must match the folder, and a one-line description of up to 1,024 characters. Siri reads only the description when deciding whether a skill applies, and reads the body only after it has matched. The body is free-form text capped at 10,000 characters. Xcode validates and compiles the bundle at build time, and the device indexes compiled skills from the app, its extensions and embedded frameworks. If two skills share a name, the system uses one of them, and which one is undefined.

The file format did not originate at Apple. App skills use a subset of the Agent Skills format published at agentskills.io, according to the documentation: the name and description fields and a Markdown body, without the scripts or reference files the full format allows. Anthropic developed the format for its Claude models and released it as an open standard in December 2025. Ad tech companies picked it up first. In July, Adform gave AI agents 29 read-only skills to query its FLOW DSP, including a platform-agnostic set built to the agentskills.io standard, and at the end of September Microsoft described a Claude Code skill, defined in a SKILL.md file, that cut Clarity beta checks from days to 30 minutes. Apple's version moves the same file into a consumer assistant that ships inside its operating systems.

Apple is clear that a skill is advice, not a program. Skills cannot be used to enforce deterministic behavior, and Siri AI does not simply run the tool intents a skill lists in order, according to the documentation. What happens in a given conversation remains the model's decision.

Testing an assistant that does not answer the same way twice

That unpredictability explains the testing tools. A Siri Integration Advisor in Xcode reviews an app's App Intents code and suggests changes. A new Siri Evaluations framework, marked as beta from version 27.3 of iOS, iPadOS, macOS and visionOS, lets developers script conversations and measure what Siri does with them.

Each scenario is a series of user turns, and each turn carries expectations: which intents Siri calls, which entities it finds, which skills it loads, what is on screen and what text appears in its reply. Every individual check is deterministic. Siri's responses are not. Apple's documentation acknowledges that answers can vary between runs and tells developers to run their sets of scenarios more than once, refining intents, parameter names, descriptions and skills until more of the set passes, and passes consistently. After each run, Xcode shows a report of what Siri said and did at every step.

Distribution is restricted while the preview lasts. Integrations can be tested on development-signed builds running on a supported device, or by internal testers in TestFlight. External TestFlight testers and anyone who downloads the app from the App Store get none of the Siri AI functionality, even if the app contains the code. Skills cannot be verified in the Simulator, only on a device.

Dates, languages and what is still missing

Apple gave two milestones without dates. "This preview provides early access before first shipping to customers in a few months; and with submission to the App Store opening up in the spring, developers have time to explore, build, and help shape their experiences before launch," the post says. The 2026 holiday season is therefore outside the window for any third-party Siri AI integration sold through the App Store.

The rules are not written yet. "As you explore early, we will follow up with more details about the terms and usage policies for experiences enabled through Siri AI. More to share soon," Apple says. The post does not indicate what those terms will cover. Developers are being invited to build, and to "help shape" the experience, before knowing the conditions under which it will be distributed.

Languages are expanding. Siri AI shipped in English on September 14 and is now available in English, French, Japanese, Korean, Portuguese and Spanish, with Traditional Chinese, Norwegian, Turkish and Vietnamese to follow "soon", according to the post. It requires an Apple Intelligence-enabled device set to a supported language, is not available to users under 13, and "some features may not be available in all regions or languages."

Capacity is rationed. Apple's footnote says certain Apple Intelligence features that rely on server-side models are subject to daily usage limits, "including but not limited to Siri AI, intelligent photo editing tools, Image Playground, and AFM 3 Cloud models in Shortcuts." It continues: "Daily limits may vary by feature, request complexity, system demand, system policies, and other factors. Expanded access to such features will be available for a fee in the future." Apple has published no number for the limits and no price for the expanded access. Nor does the post say whether a request that runs through a third-party app's tools counts against a person's daily allowance, or how a multi-step request guided by a skill is weighed against a single action.

Apple engineers are taking questions in Apple Intelligence Group Labs on November 4, from 9 to 10 a.m. Pacific time, and November 5, from 6 to 7 p.m. Pacific time.

The EU exception

One market sits outside the preview's reach for iPhone apps. "Siri AI is currently unavailable in the EU in iOS, iPadOS, and watchOS. Apple is working constructively with the EU in an attempt to find a path forward," the post says. An iPhone app built for Siri AI therefore cannot reach iPhone users in the EU, while Mac and Apple Vision Pro users in the EU have had Siri AI since September 14 when set to a supported language.

Apple has tied the iPhone and iPad gap to its reading of the Digital Markets Act, arguing that under the law a virtual assistant would have to be granted direct access to private data and control over other installed apps. At WWDC, Apple said developers in the EU would not be able to test or use the new Siri AI features for their apps on iPhone and iPad, that it had no timeline, and that the European Commission had rejected its proposal for a Trusted System Agent with an 18-month phase-in.

The Developer Preview touches that dispute directly. App Intents is the mechanism through which Siri reaches into other apps' actions and content, and in December 2024 Apple disclosed that the App Intents framework was among 15 requests for access to its technology that Meta had made under the DMA. The November 3 post opens App Intents wider for Siri. It says nothing about interoperability for assistants made by other companies. Apple's documentation does list one single-purpose schema domain, Assistant, which lets people in Japan launch a voice-based conversational app from the iPhone side button.

How Siri AI reached this point

The preview is the latest step in a sequence that began with a supplier decision. On January 12, Apple and Google announced a multi-year collaboration under which Google's Gemini models would serve as the foundation for the next generation of Apple Foundation Models, with no financial terms disclosed. At WWDC on June 8, Apple presented Siri AI, built on models developed in collaboration with Google's Gemini family, and opened it for developer testing on iOS 27, iPadOS 27, macOS 27 and visionOS 27.

Web publishers were drawn in at the same time. Apple rewrote its Applebot documentation so that crawled pages may help train Apple foundation models and supply up-to-date context when generative features produce answers, with Applebot-Extended and a nosnippet directive as the opt-outs. The asymmetry with apps is notable. A website reaches Siri AI by default unless its owner opts out; an app reaches Siri AI's custom-action layer only when its developer opts in, intent by intent.

Siri AI shipped in beta on September 14, and five languages followed in October. On October 27, Apple put an AI assistant built on Apple models and OpenAI behind its $12.99-a-month Creator Studio plan, and added MCP support on macOS 27 so that external AI agents and tools can read content, make edits and run automated actions in Keynote, Pages and Numbers. Apple now runs two different doors. Outside agents reach its own productivity apps through an open protocol. Siri reaches everyone else's apps through Apple's own framework.

Why this matters for marketers

Apple's stated aim is reach, and on the mechanics published so far, that reach favours apps people already have. Siri finds an app's content through the on-device Spotlight index, which holds content from installed apps, and acts on a background app only when the person names it. Neither the post nor the documentation pages linked from it describe Siri AI suggesting an app a person has not installed, ranking competing apps for the same request, or carrying any paid placement. In practice, the brand name a customer remembers becomes the trigger for a background request.

The words developers attach to their intents and entities now determine whether Siri matches a request, which turns titles and descriptions into a discipline close to App Store product page metadata. Paid discovery on Apple's platform still runs through App Store search, where Apple began showing additional ad positions on March 3, 2026, starting in the UK. As described on November 3, the Siri AI integration has no paid element. Whether the forthcoming terms and usage policies change that is one of the open questions.

The examples Apple chose are commercial ones. Checking a delivery, booking a reservation, placing an order, adding to a cart, choosing payment and delivery - these are agentic AI tasks completed inside a retailer's or service provider's own app, with that company's branding on the card Siri displays. Rivals have taken a different route to the same tasks. Amazon said in December 2025 that Expedia, Yelp, Angi and Square would join Alexa+ during 2026. Google's Gemini Spark agent for Mac arrived on June 30 with five new connected apps and custom MCP connections, restricted to $99.99-a-month Ultra subscribers. Meta's Muse agent gained 15 business tool connectors and access to ad accounts in the US and Canada, with nothing published, sent or spent without approval. Those assistants grow through partners and connectors that each company selects. Apple's preview is open to any developer through an operating system framework, on terms still to be published.

Behavioral data is part of the bargain. Donations tell Apple Intelligence what a person does inside an app so that the system can anticipate the next request, and the entities developers index determine what Siri can retrieve. The post and the linked documentation describe no reporting that would tell a developer how often Siri invoked its tools in production or how those requests converted. For marketers used to measuring every step of a funnel, a Siri-initiated order may arrive with little attribution attached.

The calendar matters as much as the code. Integrations cannot be submitted to the App Store before spring, the terms are unwritten, iPhone users in the EU remain excluded, and server-side use is capped at levels Apple has not disclosed. Developers have a working preview. For now, the new capabilities run in Apple's own apps and in developers' test builds.

Timeline

Summary

Who: Apple, which released the Siri AI Developer Preview to all members of its developer community, and the businesses that build iPhone, iPad, Mac and Vision Pro apps. Apple's own apps, including Calendar, Mail, Maps, Messages, Safari and Weather, are the first to carry the new capabilities.

What: A Developer Preview that lets any app expose custom actions and content to Siri AI through App Intents, using a new tool intent macro, generable entities, Spotlight-indexed content and Markdown skill files that guide Siri through multi-step tasks. Apple added a Siri Integration Advisor in Xcode and a Siri Evaluations framework for testing, and said terms and usage policies will follow. Siri AI remains subject to undisclosed daily usage limits, with paid expanded access promised for the future.

When: Apple announced the preview on November 3, 2026, with the iOS 27.3, iPadOS 27.3, macOS 27.3 and visionOS 27.3 betas. It said the capabilities first ship to customers in a few months and App Store submission opens in the spring.

Where: The preview runs on Apple Intelligence-enabled devices on four Apple platforms. Siri AI is available in English, French, Japanese, Korean, Portuguese and Spanish, but not on iPhone, iPad or Apple Watch in the EU.

Why: Until now, Siri AI could act inside third-party apps only for tasks that fitted Apple's templates. Opening custom actions turns the assistant into a channel through which customers order, book and buy inside apps just by asking, governed by metadata developers write, an index Apple controls and terms Apple has not yet published, which matters to every brand whose customers reach it through an app.