Source: Unsplash
Apple is about to let iPhone users pick their own AI model, replacing ChatGPT with Gemini, Claude, or any third-party provider via a new framework called "Extensions" coming in iOS 27 this fall. This is one of the most consequential platform decisions Apple has made in years, and from a product design perspective, it changes almost everything about how we should be designing AI-powered apps and experiences for the world's most-used mobile operating system.
I have shipped products on Apple platforms for most of my career. I've designed for iOS through the App Store era, the Siri era, the widgets era, and now the AI era. And I have to say: the iOS 27 "Extensions" framework that Bloomberg reported on May 5, 2026, is the most interesting platform shift I have seen since Apple opened Siri to third-party apps back in 2016. It is also one of the most underreported product design stories of this year.
Most of the coverage has focused on the competitive angle: Apple choosing Google and Anthropic over locking in OpenAI. That story is real. But the deeper story is a design one, and that is what I want to talk about today. When the AI layer inside an operating system becomes a user-configurable setting, everything that product designers assumed about designing for that platform becomes negotiable.
What Apple's "Extensions" Actually Does
Let me explain the feature clearly before getting into the design implications, because the details matter. Apple is building a standardized framework inside iOS 27, iPadOS 27, and macOS 27 that lets AI companies integrate their models directly into Apple Intelligence features. According to Bloomberg, users will be able to set their preferred AI provider for Siri, Writing Tools, and Image Playground by simply installing the relevant app from the App Store and enabling it in Settings.
Google Gemini and Anthropic Claude are already being tested internally. OpenAI's ChatGPT, which has been the exclusive third-party AI inside iOS since iOS 18.2 in December 2024, will move from a special-case partnership to one option among many. Apple is also reportedly releasing an Extensions API to any AI chatbot maker, so the door is open for more providers to integrate over time.
"Apple will let users choose from a range of outside artificial intelligence services to power features across its software, building on a strategy to turn its devices into a comprehensive AI platform."
— Bloomberg, May 5, 2026
There is one more detail I find fascinating from a product perspective: users will also be able to assign different voices to different AI models in Siri. So Apple's own AI might speak in one voice, while a Claude-powered response uses another. This is not just a personalization feature. It is Apple building explicit model awareness into the UX, which is a signal that they expect users to actually care about which AI is responding. That assumption changes a lot of things.
Why This Is a Product Design Problem, Not Just a Platform Feature
Here is the core challenge for product designers and app developers, stated as plainly as I can: you no longer know which AI your users are talking through when they use your app's features that tap into Apple Intelligence.
Think about what that means in practice. You build a writing assistant feature in your app. You design the UX around a certain set of AI behaviors: tone, response length, format preferences, knowledge cutoffs, capabilities. You test it, you tune the prompts, you ship it. And then your user switches their Extensions setting to a different provider, and all of those assumptions shift. The AI behavior changes. The response quality changes. Maybe the language changes. Your product experience becomes partly outside your control.
This is the same challenge every web developer faced when browsers started shipping with different JavaScript engines, but applied to the intelligence layer of the product, which is now often the core value proposition, not a peripheral feature.
via GIPHY
What This Means for Designing AI-Powered iOS Apps
I have been thinking through the specific design implications since this news broke, and I keep coming back to the same conclusions. Here is what I think every product team building on iOS needs to reckon with in 2026:
- Your AI integration strategy needs a "provider-agnostic" layer. If your app taps into Writing Tools or Siri Extensions, your prompts, your output handling, and your UI patterns all need to work across model providers, not just one. This is engineering and design work. Plan for it now, before iOS 27 ships in fall 2026. The teams that do this right will have a much smoother transition than the ones who assumed the AI layer was stable.
- Model attribution matters more than ever. When users can choose which AI powers an experience, they are going to notice when something works differently than expected. The UX needs to communicate clearly which model handled a given output. Not in a technical, jargon-heavy way, but in a clear human way: "This response was generated with Claude" with a simple logo or indicator. Trust and transparency are no longer optional design elements.
- Capability gaps become your UX responsibility. Not all models handle all tasks equally. Claude is excellent at nuanced text. Gemini has deep integration with Google's real-time data. ChatGPT has the broadest cultural recognition. If your feature works significantly better or worse depending on which model the user has selected, that needs to be surfaced, not hidden. "This feature works best with Claude" is a valid, honest UX statement. Pretending there is no difference is not.
- Fallback design becomes critical. You need to design for the experience when the user's selected model is unavailable, slow, or rate-limited. These failure states are already neglected in most apps. With a user-configurable AI layer, they become even more variable. What does your app do when the Extensions API returns an error? That has to be a designed experience, not a crash or an empty state.
- Onboarding needs to acknowledge the AI choice. If your app's value depends meaningfully on AI quality, your onboarding flow should surface the Extensions setting and give users clear guidance on which providers your app has been optimized for. This is not about locking users in. It is about helping them get the best experience from the start.
The Bigger Picture: Apple Is Building an AI Marketplace Inside iOS
Let me zoom out for a moment, because the strategic dimension of this decision is significant. Apple has, for decades, controlled both the hardware and the software stack on its devices. The App Store gave Apple enormous leverage over what experiences could reach users. This move with Extensions is Apple doing something quite different: deliberately giving up control over one of the most valuable layers of the user experience, the intelligence layer, in exchange for positioning iOS as the best platform for AI-powered products.
That is a bold call. And from what I know about how Apple has historically made platform decisions, it suggests that Apple believes it cannot win the AI model race head-to-head with Google, Anthropic, and OpenAI. So instead of trying to build the best model, Apple is building the best AI platform to run everyone else's models on. This is very similar to what Apple did with music, video, and messaging: build the best-designed container, and let the content and services compete inside it.
For product designers, this is honestly exciting. A more open AI integration layer means more design surface area to work with. It also means that the design quality of the integration experience becomes a real differentiator. A company that designs the AI choice settings, the model attribution UX, the fallback states, and the output handling beautifully is going to feel significantly better than one that slaps a generic "AI-powered" badge on a feature and calls it done.
What I Am Doing Differently in My Design Practice Right Now
Practically speaking, here is how I have started adjusting my approach when designing AI features for iOS products, given where this platform is heading:
I am designing output evaluation screens into my prototypes earlier. If a user sees AI-generated content, I want a native way for them to flag it as wrong, regenerate it, or see which model produced it. These flows are rarely designed from the beginning. They get bolted on later when users complain. They should be first-class features from day one.
I am also pushing teams to test their AI-powered features across multiple model providers early, not just the one the team uses internally. It is astonishing how often an AI feature is built and tested against GPT-4o, then discovered to produce very different output with Gemini or Claude. These differences show up in tone, length, formatting, and accuracy. Catching them in design review is infinitely cheaper than handling them after iOS 27 ships.
Finally, I am thinking more carefully about which AI tasks should be delegated to the system Extensions layer vs. which should be handled by a model the app controls directly. Not everything should go through the user's selected AI provider. Some tasks, especially the ones that are core to your product's quality promise, are better handled through a direct API integration you control. The Extensions framework is for ambient intelligence features. Your core value proposition should not be dependent on which AI model a user happened to select in Settings.
I wrote more about AI integration patterns for native mobile apps over at reloadux.com/blog. These questions about system-level AI vs. product-level AI are going to be some of the defining design decisions of the iOS 27 cycle, and they are worth thinking through carefully before WWDC on June 8.
The WWDC 2026 Watch List
Apple's WWDC kicks off on June 8, 2026. Based on everything reported so far, iOS 27 will be the most AI-focused major iOS release in Apple's history. Beyond the Extensions framework, reports point to a significantly revamped Siri with much better contextual awareness, deeper on-device processing for privacy-sensitive tasks, and performance improvements across the board.
From a product design perspective, I will be watching closely for three things at WWDC: how Apple visualizes the model switching UX, what the developer guidelines say about model attribution requirements, and whether Apple reveals any quality standards that providers must meet to be available via Extensions. Those details will shape a lot of design decisions for the next product cycle.
If you are building iOS apps right now, this is not something to wait on. The teams who are already thinking through the design implications of a multi-model AI environment will have a clear head start when iOS 27 beta drops after WWDC. The teams who treat this as a platform update to handle later are going to find themselves redesigning core UX flows in a hurry.
Are you designing for iOS 27's AI Extensions feature? What design problems are you already running into? I would love to hear what your team is thinking. Drop a comment below and let's figure this out together.
Sources: Bloomberg, "Apple Plans to Let Users Swap AI Models Across Apple Intelligence," May 5, 2026; 9to5Mac, "iOS 27 will let you choose between Gemini, Claude, and more for AI features," May 5, 2026 (9to5mac.com); TechRadar, "Apple is about to let you replace its AI with ChatGPT, Gemini and Claude," May 2026; MacRumors, "iOS 27 and macOS 27 Rumored Features," May 7, 2026; Apple WWDC 2026 scheduled for June 8, 2026