Apple's biggest developer-facing announcement in years is going to land on Monday, June 8 — and it's not a new device. It's a framework. Multiple credible reports out of March 2026 have confirmed that Apple is replacing Core ML with a modernized Core AI framework at WWDC 2026, optimized for generative-AI workloads instead of the classification and computer-vision models Core ML was originally built for.

For most indie iOS developers, this is going to feel like the SwiftUI moment all over again. Core ML didn't disappear overnight when SwiftUI shipped — but if you weren't paying attention to the framework Apple was placing its bets on, you spent the next five years writing code Apple slowly stopped recommending.

This guide walks through what Core AI is expected to be, what it doesn't change, and a concrete pre-WWDC checklist so you're not the indie scrambling to ship an iOS 27 update in October.

What Core AI actually is#

Core ML, in its original form, was an inference framework. You handed it a trained model — usually exported from PyTorch or TensorFlow via coremltools — and it ran the model locally on the Neural Engine, GPU, or CPU. That's been Apple's on-device ML story for almost a decade.

Core AI is reportedly the same idea, but redesigned around the kinds of models people are actually shipping in 2026: generative models, chat models, multimodal models, agentic workflows. The reports from 9to5Mac and AppleInsider point to a few specific architectural changes:

Foundation Models bundled in. The same on-device LLMs Apple shipped to developers via the Foundation Models framework in iOS 26 are folded directly into Core AI. You don't import a separate framework — generation, summarization, classification, and structured output are all first-class APIs.

Third-party model integration via MCP. Reports suggest Apple is building Core AI to support the Model Context Protocol, the same standard already used in agent tooling. If true, you'll be able to wire Core AI to your own model endpoint or to a hosted provider like Claude or Gemini without writing a custom adapter.

Generative-first APIs. Core ML's API surface assumed you were doing inference on a fixed input/output tensor. Core AI's reported API surface assumes you're streaming tokens, generating structured JSON, calling tools, and chaining responses — the things developers actually do in 2026.

Coexistence with Core ML. Existing .mlmodel and .mlpackage files keep working. This is a critical point: nothing shipping in your app today gets bricked by Core AI. The migration is voluntary for the foreseeable future.

What it changes for your existing app#

If your app ships a Core ML model today, the practical impact is small in the short term and large in the medium term.

Short term — your model still works. Apple has explicitly reported that Core ML and Core AI will coexist. iOS 27 won't break a 2024-era image classifier you embedded in your photo app. You don't need to rewrite anything to keep shipping the existing feature.

Medium term — Apple stops pointing developers at Core ML. Documentation gets reorganized to put Core AI first. Sample code gets updated. The optimizing compilers and Neural Engine kernels get the love. If you've been around iOS development long enough, you've seen this pattern: Storyboards still work, but UIKit best practice has been "use SwiftUI" for years now. Core ML will likely sit in the same place by iOS 28.

Longer term — features that only run on Core AI start mattering. The on-device Foundation Models, third-party model integration, and any new optimizations Apple ships will be Core AI-only. If you want the latest Neural Engine performance gains in 2027, you'll be on Core AI.

The migration timeline against the calendar#

Pinning this against the actual calendar makes it easier to plan around. The pattern Apple has run for the last five framework launches is identical:

Beta 1 (June 8–14, 2026). Core AI announced at the keynote. Documentation drops on developer.apple.com. First sample code lands. The State of the Union session usually has the most useful overview.

First-wave adoption (June 15 – July 31). Indie apps add Core AI features. TestFlight betas appear. The App Store editorial team starts watching for "designed for iOS 27" candidates.

RC and submission (August – September). iOS 27 release candidates. App Review queue gets crowded as everyone tries to ship before the public release. Reviewer tolerance for novel APIs drops.

Public release (September – October). iOS 27 ships. Apps already on Core AI get the launch-day visibility bump. Apps that wait until November are competing with everything else in a crowded "added Apple Intelligence" wave.

The indies who do well on a framework launch are the ones in the first wave. Apple has rewarded early adopters of every major framework with editorial features for a decade running, and Core AI is going to be the most editorial-friendly framework launch since SwiftUI.

What to ship before WWDC#

You can't ship Core AI integration before WWDC — the API doesn't exist publicly yet. But there's a meaningful amount of pre-work you can complete in the next 29 days that turns the post-WWDC integration from a multi-week project into a multi-day one.

1. Inventory your current Core ML usage#

Open every .mlmodel and .mlpackage in your project and ask three questions:

What's it doing? (Image classification, text classification, NLP, segmentation, regression?)
Does it need to be there? (Is the feature still earning its keep, or is it a 2022 vestige?)
Is the model size still appropriate for current device storage budgets?

You want this list ready before WWDC because Apple's first sessions on Core AI will inevitably include a "here's how Core ML maps to Core AI" slide. Knowing exactly what you have in your app makes that slide actionable instead of academic.

2. Decide your AI strategy now#

The decision tree for indie devs in 2026 is roughly:

No AI in your app. Fine. Skip the rest of this and come back when you have a feature that wants it.
Pure on-device AI (no network calls). Foundation Models and Core AI is your path. Privacy story is clean. App Review loves you.
Hybrid AI (on-device for privacy-sensitive stuff, cloud for heavier workloads). Core AI plus a hosted model. This is where most indies will land.
Full cloud AI. You don't strictly need Core AI for this — but the framework will likely make the on-device fallback path easier than rolling your own.

Pick one. The strategy determines what kind of WWDC sessions you watch and what kind of Core AI APIs you care about.

This is the boring one that bites you in review. Apple's 5.1.2(i) AI consent rule, which started getting enforced earlier this spring, requires apps that send user data to AI models to disclose it clearly and get explicit consent. Core AI integration is going to bring this to the front of the queue for a lot of apps that currently don't think of themselves as "AI apps."

4. Fix your store listing now#

Your existing store listing was written for whatever your app does today. The minute you ship a Core AI feature, you're going to want to mention it. The minute you mention it, App Review's editorial team is going to hold the listing to a higher standard than they did six months ago.

Concretely: review your app description for vague AI claims, missing disclosures, and any reference to features you don't actually ship. Apple's editorial team has been rejecting listings for unsubstantiated AI language since March, and the bar will only get tighter as iOS 27 approaches. "AI-powered" without specifics is a flag. "Uses Apple Intelligence to summarize your notes" with a corresponding privacy disclosure is fine.

5. Get your screenshot pipeline ready for the iOS 27 UI#

Every iOS major release breaks somebody's screenshots. Status bar layout changes. New control center. Adjusted safe areas. New system UI elements that show up in the background of your captures. If your screenshot pipeline is "I open Sketch and update them by hand," you're going to spend the first week of iOS 27 beta redoing screenshots instead of writing Core AI code.

The fix is automation. Whether that's fastlane snapshot, your own SwiftUI render pipeline, or Stora's screenshot generator, the goal is the same: when iOS 27 ships and your screenshots are wrong, you regenerate the whole localized set in minutes, not days.

The pre-WWDC checklist#

Five concrete items, each one a few hours of work, all doable in the next 29 days:

List every Core ML model in your app. Name, purpose, size, last updated.
Pick your AI strategy. None / pure on-device / hybrid / full cloud.
Audit your privacy strings and consent flow against 5.1.2(i). If you ship anything generative, you need explicit consent.
Run your store listing through compliance. Strip vague AI claims. Make sure every feature you mention actually exists in the build.
Confirm your screenshot pipeline can regenerate the full localized set in under an hour. If it can't, fix it before iOS 27 beta drops.

Do these five things before June 8 and you're ahead of 90% of the indie field on Core AI day.

The bigger picture#

Core AI is the framework. The change underneath it is that Apple is finally treating on-device AI as table stakes for app development, the same way it treats networking or storage. Foundation Models in iOS 26 was the appetizer. Core AI in iOS 27 is the main course.

The indie devs who do well in this transition won't be the ones with the fanciest model integrations. They'll be the ones whose submission pipeline doesn't bottleneck on screenshot redos, listing rewrites, or compliance flags. Core AI integration is a few days of work. The submission churn around it is what eats the calendar.

Stora was built specifically for that bottleneck. The screenshot generator regenerates localized sets when iOS 27 inevitably breaks them. The AI store listing rewrites your description against current iOS metadata rules so an editorial reviewer doesn't reject you for the AI claim you wrote in 2024. The compliance engine flags 5.1.2(i) issues before review does. None of that helps you write better Core AI code — but it gives back the days you'd otherwise burn on submission churn during the most important framework transition Apple has run in years.

29 days. Use them.

Core AI Is Replacing Core ML at WWDC 2026: The Migration Plan for Indie iOS Devs