Apple Is Figuring Out How to Put AI Agents in the App Store

Why AI Agents Don’t Fit the App Store Model

Apple is actively exploring how to incorporate AI agents into its App Store, The Information reported this week — trying to benefit from the agent wave without relinquishing the platform control that defines its entire business model. The tension in that sentence is the whole story.

Traditional iOS apps are discrete, sandboxed units. They have a bundle ID, go through review, and live in the App Store. You install them. You open them. They do a thing.

AI agents are categorically different. An agent is a process that reasons, plans, and takes actions — often across multiple tools, APIs, and data sources — without a human clicking through every step. A capable agent might draft a reply in Mail, pull context from Calendar, call a web API, and update a CRM record, all from a single prompt. That’s software behavior, not a software product. There is no icon for it.

Apple’s existing frameworks give developers the right hooks for this. App Intents, introduced in iOS 16 and extended significantly in iOS 17 and beyond, let apps expose their core actions to external orchestration — including Siri and third-party agents. Siri Extensions, which shipped with iOS 27, push this further. But those are interfaces to apps, not agents themselves. A full-stack agent that coordinates multiple apps and external services doesn’t map to any existing App Store category.

The Revenue Model Problem

The App Store’s 30% commission works because Apple controls the only legal distribution channel for native iOS apps. If you want to reach iPhone users with a native app, you go through the App Store. That’s how the toll is collected.

AI agents threaten to route around that chokepoint in two ways.

First, the most capable agents are hosted services, not iOS binaries. A user signs up on the web, connects their iPhone via a lightweight companion app, and the agent runs in the cloud. The iOS app — if there is one — is little more than a permissions bridge. Apple collects 30% of the companion app’s subscription, which may be a fraction of what the agent costs to run.

Second, the pricing model for agents is wrong for the App Store billing system. Subscriptions and one-time purchases work well for traditional apps. Agents are increasingly priced by token usage, API calls, or task completions — metered consumption that doesn’t translate to a recurring App Store subscription. Asking Apple to take 30% of each LLM API call an agent makes would be absurd. But under current rules, that’s approximately what the logic implies.

Apple’s Actual Options

Apple has a narrower set of moves than it might appear. Too restrictive, and iOS becomes the laggard platform for AI development while the interesting agent products go to Android and the web. Too permissive, and it loses the enforcement layer that makes the App Store valuable in the first place.

The EU’s Digital Markets Act adds further constraint: it compels Apple to allow third-party app distribution and meaningful interoperability with iOS features. A policy that effectively limits AI agents to Apple’s own ecosystem would draw regulatory attention quickly.

What Apple is reportedly working through is how to create a category-specific framework — the same pattern it has used for every prior disruption that didn’t fit the existing model. Based on how Apple has handled web apps, extensions, and widgets, a workable structure would probably include:

A dedicated App Store category for agent apps with distinct review criteria
New entitlement tiers for agent-level capabilities: reading across apps, taking actions in the background, accessing sensitive data on behalf of the user
Mandatory disclosure UI standards so users know when an agent is acting autonomously
Some form of revenue participation on agent subscriptions brokered through App Store billing, with carve-outs for usage-based plans

None of that is official. But it matches how Apple has historically handled things that mattered too much to ignore and fit too poorly to absorb unchanged.

What to Do Right Now If You’re Building AI Features on iOS

The policy landscape is unresolved, but the architectural decisions are not. Here is what we would do today.

Build your agent capabilities through App Intents. This is the sanctioned interface for exposing your app’s actions to external orchestration — including Siri, Shortcuts, and any agent framework Apple formalizes. Structuring your core actions as App Intents means you’re positioned for whatever comes next rather than scrambling to retrofit it.

Keep orchestration logic in your backend, not in the iOS binary. If the agent’s reasoning layer lives server-side, you can update its behavior without triggering an App Review, and you can serve the same agent across platforms. The iOS app should be a capability provider — managing permissions, surfacing context, executing on-device actions — not the agent itself.

Design for explicit user consent on every autonomous action. Any agent capability that sends a message, makes a purchase, books something, or modifies user data will eventually require a defined consent flow. Build that affordance into your UX now. Apple has required this in every prior case where apps acted beyond immediate user input, and agents are a more extreme version of the same pattern.

We’ve navigated similar inflection points on Amali and TeleTabeb — products where the underlying platform APIs were still settling while the product needed to be stable. The principle is the same: build to the stable interface (App Intents, entitlements, explicit consent), not to the future policy state.

What We’re Watching

WWDC 2026 is weeks away. It is the natural venue for Apple to lay out a formal agent framework — possibly an extension of the Siri Extensions system from iOS 27, possibly something new at the OS level. Watch for new entitlement types, an “Agent” category in App Store Connect, or changes to the StoreKit billing APIs that accommodate usage-based pricing. Those will signal how Apple intends to monetize and govern this layer.

The deeper question is whether Apple can maintain platform control as AI agents commoditize the UI layer users interact with. The App Store was the right answer for the app era, when the product was a discrete, installable thing. Agents are a different unit of software entirely. Apple knows this, which is why they’re exploring it now rather than after the market has already formed around something they don’t control.

Sources

Apple Explores Ways to Welcome AI Agents in the App Store — The Information, May 13, 2026
App Intents framework — Apple Developer Documentation
App Store Review Guidelines — Apple Developer