On June 8, 2026, Tim Cook walked off the WWDC stage for the last time as CEO, and what I remember from the keynote isn’t Siri AI or macOS Golden Gate. It is a line from the Platforms State of the Union the next morning: apps recompiled with Xcode 27 adopt Liquid Glass automatically, and the opt-out is gone. That is the whole story of iOS in 2027 in one sentence. Apple has stopped asking. The platform now sets the defaults, and the apps that win next year will be the ones built on those defaults instead of fighting them.
I lead mobile work at LITSLINK, a US-based software company with iOS teams in Palo Alto, Orlando, and Europe. We have shipped iPhone apps since 2014, and I have never seen a twelve-month window where so many Apple decisions landed at once. Below is what we think takes off in 2027, why, and what we are telling clients to do this quarter.
The on-device default, and the consensus it replaces
Our name for the pattern is the on-device default. The first question for any new iOS feature is no longer “which API do we call” but “does the phone already do this for free, and how do we wrap it.” Apple’s roughly 3-billion-parameter on-device model, the Foundation Models framework around it, the App Intents schemas that feed Siri and Spotlight, the Liquid Glass materials that now apply themselves: all of it pushes compute, context, and visual language down to the device and out of your budget.
The lazy consensus runs the other way. I hear a version of it in almost every discovery call:
“Apple Intelligence makes app features free now. Summaries, rewriting, search, all of it is built into iOS. We don’t need a native team for that anymore; we just need a cross-platform shell and the OS does the rest.”
Half of that is true, which is what makes it dangerous. Inference is free. Integration is not. The Foundation Models API is a Swift API, guided generation works through the @Generable macro on Swift types, tool calling means implementing a Swift protocol, and Siri’s new View Annotations only see what your SwiftUI or UIKit views expose. A cross-platform shell gets the generic OS features that every app gets. It does not get the parts that make your app show up when a user asks Siri for it. The “free” features are free for your competitors too.
Six trends that take off in 2027
- On-device inference becomes the architecture, not the feature. At WWDC25 (June 9, 2025), Apple opened its on-device model to any app, and in the 2025 technical update it described that model as a compact, approximately 3-billion-parameter network compressed to 2 bits per weight, with 15 supported languages (Apple Machine Learning Research, updated July 17, 2025). A year later, Apple added image input, let the same Swift API call third-party server models by name (Claude and Gemini were the two mentioned), and gave developers with fewer than two million first-time App Store downloads no-cost access to models on Private Cloud Compute. For most apps, inference now costs nothing until you are already big, so the cloud becomes the place you go only for what a 3B model cannot do. We restructured two client backends in Q3 2026 around that split, and both LLM line items dropped to near zero.
- Siri turns into a distribution channel, and apps that are not indexed stop existing. The new entity and intent schemas announced on June 8, 2026 let apps contribute content to Spotlight’s semantic index, and View Annotations let Siri act on what is on screen. Combine that with Siri AI reading Messages, Mail, and Photos at the OS level, and the user asks for an outcome and iOS picks the app. My inference, not something Apple said: in 2027 “App Intents coverage” becomes a growth metric the way “App Store screenshots” was in 2019. The apps that expose their entities get picked. The rest get a tab in a folder.
- Liquid Glass stops being optional, and legibility becomes a product metric. iOS 26 shipped the new material with a compatibility opt-out. Xcode 27 removed it. Every app rebuilt with the new toolchain takes the design, period. Apple also added a user-facing transparency slider in iOS 27 that runs from ultra-clear to fully tinted, which is Apple quietly admitting the first version hurt readability. Your UI now has to hold up at both ends of that slider, on top of user photos, in light and dark mode. We test every screen at three slider positions. It is tedious. It also caught four contrast failures in one fintech app before a single user saw them.
- Swift 6 strict concurrency moves from “someday” to “blocker.” Swift 6.2, released September 15, 2025, added the main-actor-by-default isolation mode and the @concurrent attribute, which together cut most of the pain out of adopting the Swift 6 language mode. Swift 6.4, shown at WWDC26, added async support in defer and better type-checker diagnostics. Excuses are running out. More to the point, the new Foundation Models and Core AI APIs are async-first, so an app still in Swift 5 mode with a pile of unaudited DispatchQueue calls cannot adopt the 2027 features cleanly. Migration is the price of admission.
- SwiftUI becomes the only sane default for new screens. Apple said layouts resize up to twice as fast in the new SwiftUI, added reorderable containers and swipe actions for any container, and used Notion’s move of its UI to SwiftUI as the headline example. I was a UIKit holdout longer than most of my team, so take this from a convert: the remaining UIKit arguments in 2027 are camera pipelines, heavy custom gestures, and legacy code you have not had time to replace. Everything else is SwiftUI.
- The EU splits every iOS roadmap in two. On August 18, 2026, Apple rewrote its EU business terms again, effective October 1, 2026: a 5% Core Technology Commission on purchases in apps from alternative stores or the web, 15% on App Store apps that link out, and a waiver for small marketplaces. Siri AI, meanwhile, will not ship on iOS or iPadOS in the EU or in China this fall. If you sell in Europe, you will be maintaining two feature flags, two payment flows, and two sets of store assets in 2027. Not forever, maybe, but for the whole year.
Build now or wait: the six trends ranked by maturity
| Trend | Maturity as of October 2026 | Build now or wait | What makes it risky |
| On-device Foundation Models | Shipping since iOS 26, image input and third-party models added in iOS 27 beta | Build now | 3B model limits: short context, no deep reasoning, 15 languages only |
| App Intents and Siri View Annotations | New schemas in iOS 27 beta, Siri AI is a waitlisted beta | Build the intents now, wait on Siri-specific UX | Siri AI absent in the EU and China, usage allowances still undefined |
| Liquid Glass adoption | Mandatory with Xcode 27 | Build now, no choice | Legibility across the transparency slider |
| Swift 6 strict concurrency | Stable since Swift 6.2, Sept 2025 | Build now | Migration effort on large UIKit codebases |
| SwiftUI as the default UI layer | Mature, with 2x layout gains in iOS 27 | Build now for new screens | Camera, custom gesture, and legacy parity gaps |
| Resizable iPhone apps on iPad and Mac | Automatic opt-in on rebuild | Prepare, do not redesign yet | Foldable iPhone is speculation, not an announcement |
The last row deserves a flag. Apple said iPhone apps on iPad and in iPhone Mirroring can now resize, and that apps opt in automatically when rebuilt. Plenty of people read that as groundwork for a folding iPhone. That reading is an inference, and nobody at Apple has confirmed any device. Build adaptive layouts because they are correct, not because of a rumor.
The numbers behind the bet
Trend pieces are cheap. Here is what we are anchoring on.
- Apple’s App Store study by Analysis Group, published June 4, 2026, put developer billings and sales at over $1.4 trillion for 2025, with more than 90% of that carrying no Apple commission. The same release says more than 40 of the top 100 apps in 2025 featured consumer-facing AI, and apps with AI features saw four times the billing growth (Apple Newsroom, June 4, 2026).
- Apple’s own App Store measurement on February 12, 2026, 150 days after launch, showed iOS 26 on 74% of iPhones introduced in the last four years and 66% of all iPhones. iOS 18 was at 76% and 68% at a similar point. The install base moves to the new OS within a few months, every year, which is why “wait for adoption” is not a real strategy on this platform.
- In the 2025 Stack Overflow survey, 5.4% of all respondents and 5.7% of professional developers reported using Swift, and 51.9% of those who used it said they want to keep using it (Stack Overflow Developer Survey 2025). Kotlin sat at 10.8% and Dart at 5.9%. Swift is a small, loyal pool. Hiring for it in 2027 will not get easier.
- At WWDC26 Apple said Xcode 27 is about 30% smaller, Apple silicon only, and that Xcode Cloud builds run up to twice as fast. Our own CI numbers on the beta are closer to 1.4x on a mid-size app, but the direction holds.
One more number, from our side. In the two 2026 projects where we moved summarization and classification from a hosted LLM to the on-device model, we traded a monthly inference bill for roughly three extra weeks of engineering to handle the shorter context window and the fallback path for unsupported devices. Both clients took that trade. Whether it generalizes to your app depends on how much of your AI workload fits inside a 3B model’s head, which is worth two days of prototyping to find out.
What to look for in an iOS team in 2027
Hire dedicated iOS developers who have already lived through a Swift 6 migration, not ones who plan to learn it on your codebase. At LITSLINK, our vetted engineers work in Swift and Objective-C across the iOS SDK, Apple frameworks, and AI and ML integration, with US-hours project management and senior European engineering behind it. Clients interview pre-vetted candidates by video call and can start in as little as 48 hours, and if a developer is not a fit we replace them at no cost. The 2027 checklist is specific: Foundation Models and App Intents experience, SwiftUI as the default, Liquid Glass testing habits, and someone who can argue with you about actor isolation.
Yevhenii Melikov, Machine Learning Engineer at LITSLINK, shared a comment for this piece on the on-device point: “The first thing we learned with the Foundation Models framework is that a 3 billion parameter model at 2 bits per weight is not a cloud model with a smaller bill. It is good at summarizing and extracting from text the user already has, and vague the moment you ask it to reason beyond that. So we moved the open-ended steps to a cloud model and kept the on-device one for the structured work, which is where the guided generation macro really pays off. The headache was not the model; it was making sure every call into it was already async and actor-isolated, which is exactly what Swift 6 forces on you.”
That matches what we saw on the Swift 6 migrations this year: more warnings than expected, most of them legitimate, and a payoff that only shows up when the AI features go in.
What to do now, in order
- Turn on Swift 6 language mode in a branch and count the warnings. Do not fix them yet. Count them, group them by module, and put a number of sprints next to each group. That single afternoon turns a vague “we should migrate” into a budget line.
- Prototype your one most expensive AI call on the on-device model. Take the feature with the biggest monthly inference bill, rewrite it against Foundation Models with a @Generable output type, and run it on an iPhone 15 and an iPhone 17 Pro. If the quality holds, you have found money. If it does not, you have learned where the 3B ceiling sits for your use case, which is worth knowing before 2027 budgets close.
- Define your App Intents and entities before you design new Siri UX. The schemas are the part that survives whatever Siri AI turns into. The conversational layer is still a waitlisted beta with daily allowances, and it is not shipping in the EU. Index first, converse later.
- Rebuild on the Xcode 27 beta and walk every screen at three transparency slider positions. Note contrast failures and text over photos. Two days of work for most apps, and it prevents the one-star “I can’t read anything” reviews that followed iOS 26.
- Fork your EU roadmap on paper. List every feature that depends on Siri AI, every purchase flow, and every store asset, and mark which version ships where after October 1, 2026. If the fork costs more than the EU revenue, fine, but decide that with the list in front of you.
- Pick the three UIKit screens that move to SwiftUI first. Settings, onboarding, and any list-heavy screen are the usual winners. Leave the camera alone.
Where this is heading after 2027
My honest expectation, labeled as such: by WWDC27 the on-device model gets a larger sibling that lives on Private Cloud Compute with the same Swift API, and the line between “on-device” and “cloud” stops being a developer decision and becomes a runtime one. The Dynamic Profiles system for multi-agent workflows that Apple showed in June 2026 reads like the first step toward that. Nothing about a bigger on-device model has been announced, so treat that as a forecast with a decent track record behind it and nothing more.
The part I am more sure about is structural. Apple has spent two years turning features into platform defaults: AI inference, visual design, Siri integration, even Mac-only binaries. Each default removes a decision from the developer and adds a requirement. The teams that treat those requirements as the spec, rather than as constraints to route around, will ship faster in 2027 and will be the ones Siri recommends. Our team at LITSLINK has placed that bet with our own hours.
Your iPhone stopped being the client a while ago, and 2027 is the year the apps catch up.
