On the latest episode of Upgrade, Jason Snell and Myke Hurley covered an important part of the Apple Intelligence beta that I’d like to highlight. Leading up to WWDC, Daniel and I talked on Core Intuition about how great it was going to be to have a small on-device LLM available for developers to use. Why pay to send requests from your app out to OpenAI or Anthropic (or even run your own servers with Llama) when you can just use Apple’s model directly?
But that hasn’t happened. I’ve transcribed the relevant segment from Upgrade discussing this, included below. There’s more before and after that is good too.
Jason: The fundamental purpose of a developer beta is supposed to be for developers to use new features in order to plan their release for when that version comes out. That’s what a developer beta is for. That’s why it’s called a developer beta. And the problem with it is, developers can’t do much with Apple Intelligence. The big thing they can do is the Intents stuff, which isn’t in there.
Jason: It struck me… So you’re talking about Apple Mail and summarization. What Apple hasn’t done is make — maybe this is why it’s not going to be available in the EU — what they haven’t done is make an API so if you’re the developer of…
Myke: Yep.
Jason: …let’s say Mimestream…
Myke: Slack.
Jason: …or anybody. That there’s not like, I’m going to hand a message to Apple’s LLM and ask for a summary. I’m going to go to SummaryKit, or whatever, hand them this information, get a summary back and put it in my UI. No, it’s just in Mail. And so I think that’s one of the frustrations that I’ve got with the way Apple is sort of saying “it’s for developers” and all that, because a lot of these features… And I understand why. I realize it’s early days yet. But Apple Intelligence, a lot of these features are only going to be useful if you’re using the stock Apple apps. And I don’t love it.
Jason: And I’m sure that in the fullness of time there will be APIs for third-party developers to use that will give them access to the same kind of model and summation and all of that that Apple uses, but I don’t believe any of that is available right now.
I’m very puzzled by this omission too because it seems like the most basic low-hanging fruit to open up for developers. There’s an LLM on the phone. Let developers use it. It could be as simple as a single API call to pass the prompt and text to process. This must be there already as a private API that Mail and Messages are using.
Perhaps Apple is worried about how they are scaling the private cloud compute and don’t want developers to use anything that might touch it. Perhaps Apple Intelligence is limited to so few devices that it would create confusing minimum hardware requirements for third-party apps. Perhaps Apple never wants to give developers direct LLM access because Apple considers Apple Intelligence a user feature, leaning on third-party data without developers having any control over AI features in their own apps.
Who knows. But it means developers will need to stick with OpenAI and similar APIs for at least another year or longer. Apple has a unique architecture for AI. I like the potential, even if their strategy seems mostly about their own users. If there will be innovation with AI on iPhones, though, only Apple can do it. Meanwhile the rest of the tech world continues full-speed ahead with a more developer-focused approach to AI, with many platforms to build on.
