Some takeaways from Apple’s announcements at WWDC!

• It was almost official-official that LLMs are pure commodities, but Apple supporting even the very minimal, almost cosmetic integration with OAI GPT / Google Gemini suggests they’re hedging against that bet. If someone launches an LLM that is meaningfully and sustainably differentiated, Apple users can “plug into it,” to some extent. But it seems pretty clear that Apple thinks LLMs are mature, and, to some extent, generic.

• These kinds of features really do make the most sense at the OS level, in my view. The technologies will be important for many companies in countless, unimaginable ways, but for users to leverage LLMs, ubiquity / availability, consistency, and safe access to personal data are all so important that I think it’s defensible from a “principled” perspective that OS-makers are going to own a ton of this. But it’s also a matter of “who has leverage,” and often it’s still OS-makers.

  • One wrinkle I’m curious about is: how often will Apple update the models powering all this? Is it an OS-cycle cadence, annually? That seems way too low, especially assuming that they’re likely behind the state of the art already, substantially in some areas, and need to catch up! Apple’s “image maker” cannot be dramatically worse than others for 11-12 months at a time, can it? (Maybe it can!).

• No one believes me when I say this, but every single Apple event in history has been met with boos and jeers and shrugs and critical thinkpieces. If you remember otherwise, you’re misremembering! I often think that the worse the reaction, the better the launch, because verily I say unto you: no one knows shit, and the opinions of people who don’t know they don’t know shit —and who thus share their takes, as I do here— are especially bad indicators of what “most people will think.” I bet people are going to absolutely love this iOS update in particular. Fun customization / aesthetic stuff plus generative / filtering AI doing crazy stuff?! For almost all iOS users, this will be the first time they’ve ever seen this sort of thing! People will probably dig it. If the Photos redesign is good, it’ll be a party on Main Street (has a redesign ever been good?! of an app your oldest relatives depend on and love?! psychotic).

• I think A-team is now on Vision, B-team on iOS, etc. You can sort of tell just from the UI design, setting aside product strategy entirely. The Vision updates all look good (still not getting one).

I suspect this was some of the fastest work Apple has ever done, and I’m surprised to admit they must have been panicked to some extent, most probably fearing pressure from Microsoft in particular. People will take for granted how hard it probably was to

• take all the best practices from the world of AI / LLMs and make them work “on device” / with “private cloud” computing (and make it all efficient); probably some gnarly true innovation / tradeoffs in that;

• structure it all with iOS so that it can work with your apps, personal data, screen elements, etc., but not slosh your personal information or other sensitive data all around (Apple has crushed on security and I’m sure they don’t want to fumble);

• jam this insane modal-conversation-interactive set of UIs into dozens of different spaces and make it even halfway sensible (I swear some of these screens look like they must be mock-only / unbuilt because they’re just TOO WEIRD!!!);

and esp. within the mode Apple normally works in! They were hauling, seems like to me anyway.

For Substack, I think a few things are really salient:

• They put tabs in Mail. They’re going to Gmail it. Email deliverability is going to be hurt.

• They’re using AI on notifications. Notification deliverability is going to be hurt eventually. This may be a blessing in disguise: forcing us to ensure that notifications are really valuable / desirable.

• They’re changing some contact-matching stuff that may hurt some network growth.

Obviously, there doesn’t appear to be any need for Substack to work on our own generative models or even integrations of generative models into our input fields; it would be premature to conclude this now, probably, but within a year we may see a very large % of the interested market using devices that have their own such features incorporated. We’ll see!!!