Apple spent most of 2023 and 2024 getting roasted for missing the AI boom. Google was stuffing Gemini into every product it owned, Samsung printed “Galaxy AI” on the box, and Apple’s answer at WWDC in June 2024 was a name, Apple Intelligence, plus a promise that most of it would run on the phone in your pocket rather than in a data centre.
Two years on, the picture is clearer. A few parts work beautifully. A few parts are still embarrassing. And a surprising amount of Apple AI has quietly slipped into hardware you already own: the watch on your wrist, the earbuds in your bag, the hinge on a phone that folds.
Here’s what it actually does today, where it’s genuinely good, and where it still falls apart.
Apple Intelligence, in plain English
Apple didn’t ship a chatbot. It shipped a layer of features baked into iOS, iPadOS, macOS, watchOS and visionOS. If you own an iPhone 15 Pro or anything newer, or an M-series iPad or Mac, you already have access to most of it. You just may not have switched it on.
The pieces people use day to day look like this:
- Writing Tools — rewrite, proofread or summarise any selected text in Mail, Notes, Pages and most third-party apps.
- Notification summaries — a busy group chat or a 40-message thread compressed into two lines on your lock screen.
- Clean Up in Photos — tap a stranger photobombing a holiday shot and the model fills in the background behind them.
- Genmoji and Image Playground — generate a custom emoji or illustration from a written description.
- Visual Intelligence — point the camera at a restaurant, a plant or a poster and ask what you’re looking at.
- Live Translation — real-time translation in Messages and FaceTime, plus a conversational mode through AirPods.
None of that is dramatic. It’s the pattern Apple tends to follow: arrive late, ship something dull-looking, then make it the default behaviour of a billion devices.
The privacy plumbing is the real engineering story
Apple runs two models. A roughly 3-billion-parameter model lives on your device and handles the small stuff: summarising a paragraph, tagging a photo, sorting an email. When a request needs more horsepower, it gets handed to Private Cloud Compute, which runs on Apple silicon servers built specifically for this. Apple says your data isn’t stored, and that independent researchers can inspect the software images to verify the claims.
That’s a meaningfully different bet from the cloud-first approach most rivals took, and it’s also why the feature list grows more slowly than Google’s. Running inference on a phone is harder than running it in a warehouse full of GPUs.
The same instinct runs through the rest of the product line. If you’ve ever wondered how to tell if someone else’s AirTag is tracking you, that warning system is Apple building privacy into hardware rather than into a settings menu.
One practical detail: Apple Intelligence wants about 7GB of free storage, and it will happily sit in “downloading” for an hour on a slow connection.
Siri is finally getting the rebuild it needed
Siri has been the punchline of the assistant world since roughly 2016. The 2024 refresh changed its look, a glowing rim around the screen edge, and let you type to it instead of talking. Useful, but cosmetic.
The overhaul that matters is on-screen awareness and personal context: Siri reading your messages, calendar and mail to answer things like “when is my sister’s flight landing?” It slipped. Apple confirmed in March 2025 that the work was taking longer than expected, and the personal-context version is now expected in 2026.
That delay stings, because it’s the piece that turns Siri from a novelty into something you’d trust with real work. It’s also part of a wider shift, since smart assistants are quietly moving from answering questions to doing the tasks themselves, and Apple can’t afford to sit that one out.
The Apple AI you’re already wearing
On your wrist
The most useful AI Apple shipped recently isn’t in a phone at all. When Apple unveiled the Watch Series 12 and Watch Ultra 4 with an AI upgrade that recaps your day, the headline feature was a short written digest of what you actually did: workouts, meetings, a walk you’d forgotten about. It reads like a gimmick until you start relying on it.
In your ears
AirPods have quietly become Apple’s most interesting AI input device. Live Translation handles a two-way conversation with someone who doesn’t share your language, and the hearing health features turned a pair of earbuds into a clinically validated hearing aid. If you’re weighing an upgrade, it’s worth checking where to preorder the Apple AirPods 5 before launch stock disappears.
Inside the factory
Some of Apple’s AI work never reaches a user interface. The hinge for Apple’s new foldable phone was built with AI, with generative models used to explore designs and machine vision checking tolerances during assembly. Unglamorous, and arguably the highest-value AI Apple runs.
How it stacks up against Gemini and Galaxy AI
All three big phone makers now ship AI. The differences are philosophical:
- Apple defaults to on-device processing, asks permission before sending anything to a server, and doesn’t tie features to an advertising business.
- Google leans cloud-first, integrates deeply with Workspace and Search, and moves fastest on new capabilities.
- Samsung mixes both, gives you per-feature toggles, and has been quicker to push AI onto mid-range hardware.
If you want the newest tricks first, Google wins. If you want the AI least likely to surprise you in a bad way, Apple’s caution has real appeal, though that same caution is why Siri feels a generation behind.
Where it still falls over
Notification summaries misfired badly. In January 2025 the BBC complained after a summary invented a headline about a murder suspect, and Apple paused the feature for news and entertainment apps before bringing it back as an opt-in. A small embarrassment, but it revealed something bigger: summarising someone else’s words incorrectly is a far worse failure than generating a weird picture.
There are other gaps. Apple Intelligence arrived late in the EU after regulatory wrangling. Siri still struggles to chain two actions together without losing the thread. The on-device model gets confused by unusual names and niche jargon. And the ChatGPT handoff, useful as it is, feels like a plug rather than a design.
How to switch it on
Open Settings, tap Apple Intelligence & Siri, and flip the toggle. You’ll sit on a brief waitlist while the models download, and you’ll need those 7GB of free space.
Then try three things today: summarise a long email thread you’ve been avoiding, run Clean Up on a photo you thought was ruined, and switch on notification summaries for a group chat you’d rather skim. If none of those feel useful, you’ve learned something about where the technology actually is.
What’s coming next
The next twelve months decide whether Apple AI is a footnote or a foundation. The personal-context Siri build lands first, assuming it doesn’t slip again. App Intents should let third-party apps hand Siri real actions instead of just opening a screen. Apple has also signalled that ChatGPT won’t be the only outside model on offer, with Gemini widely expected to appear as a choice.
Meanwhile the hardware keeps absorbing the work. On-device models improve every time Apple shrinks a chip, and every iPhone with more memory makes the cloud less necessary.
Apple’s version of AI will probably never be the flashiest one you can buy. It’s aiming to be the one you stop noticing, quietly rewriting your email, flagging the thing you missed, and keeping the data on your side of the fence. Whether that’s enough depends on whether the next Siri shows up on time.