If you’re a Coder Radio listener, you already know the deal; I’ve spent a lot of time over the years criticizing Apple’s platform decisions and their generally imperious attitude toward their developer community. Whether it’s their heavy-handed sandboxing, the App Store tax, or hardware choices that made developers flee to Linux and System76 machines, I haven’t exactly pulled my punches, but credit where credit is due. Apple’s latest WWDC announcements around Core AI aren’t just the usual reality distortion field. They’ve actually managed to build an AI strategy that is genuinely good for users (championing privacy) while simultaneously throwing a massive lifeline to indie devs. There’s plenty of details still to be revealed, but let’s hit the main points.
1. Free Compute Is Best Compute
Developing software is expensive. Adding AI into the mix usually makes it exponentially worse. Running modern AI features typically means renting expensive GPU servers or racking up large API bills from the likes of OpenAI who can (and likely will in the not-so-distant future) raise their rates whenever they like. Apple’s push for “on-device first” inference with Core AI changes the math completely. It turns the user’s iPhone, iPad, or Mac into the compute engine. For a solo dev or a small dev shop, that means free compute. You get to add powerful, cutting-edge generative AI features to your app without a recurring cloud bill hanging over your head.
Better yet, because the data never leaves the user’s local hardware, you get to slap a “Privacy-First” badge on your App Store page, assuming this isn’t some sort of violation of the many byzantine App Store rules. In an era where users are increasingly paranoid about their personal data being scraped and sold (and rightly so), that is a hell of a selling point. You get to ride the coattails of Apple’s multi-billion dollar privacy marketing campaigns, and it doesn’t cost you a dime in server fees; in fact, you’re likely to at least for the first year or so of a moderately successful app save money.
2. PyTorch Will Light Your Way
Instead of doing the traditional Apple (at least in the age of iOS) thing of eschewing an industry standard in favor of a proprietary technology, Apple has surprisingly embraced PyTorch. With the introduction of Core AI PyTorch Extensions, you don’t have to rebuild your machine learning models from scratch. You can take the open-source PyTorch models that you (or, more likely, the broader community over on Hugging Face) have already built in Python and convert them straight to Apple’s native .aimodel format. This is welcome practicality; it saves hours of frustrating translation work, and it shows that Apple is actually paying attention to how developers are doing ML out here in the real world. You can build in the open-source tools you love and deploy to Apple’s silicon with little hassle.
3. Cloud Compute Subsidies for Small Businesses
Of course, not everything can run locally on a phone. Sometimes you just need the dual-jet engine horsepower of a pod-racing server foundation model. Normally, this is where the indie developer gets priced out, unable to compete with the VC-backed giants who can afford massive cloud infrastructure. But Apple’s new Private Cloud Compute (PCC) offering comes with a massive perk for small businesses. If you’re enrolled in the App Store Small Business Program and your app is under the 2-million download cap, Apple is basically waiving the API fees to hit their server-side Foundation Models. The beauty here that Apple is footing the server bill.
If you’ve ever been a first-time indie dev, you know those server-costs can start to bite quick, especially if you’re not experienced in back-end development or database optimizations / administration. Of course I still love my Thelio and the movement to shift to desktop Linux for development, but it’s nice to see Apple finally releasing something that will genuinely provide novel opportunities for the indie development community. Let me know what you think on Coder Radio or LinkedIn. Also, if your higher-ups are demanding that you bring “AI” into your workflows but you have tons of legacy data and processes, take a look at this; Hell, you can even just forward it to them and have them take the quiz.
