Steve Troughton-SmithIt feels *crazy* to me that a company as conservative as Apple is letting Claude loose on the source code to its frameworks and apps. I know @markgurman has been reporting on it, but I've heard the same from a couple of different places now. I'd be fascinated to know what percentage of projects in the next set of OS releases from the company will have been touched by it? If you use anything with a screen, there's just no escaping AI anymore (and you're kidding yourself if you think otherwise)
("It's not like quality could get any *worse*…", you say)
Paul Hudson@stroughtonsmith If Claude makes AirDrop work 100% of the time and nothing else, I'll be happy ✨
QuarterSwede@twostraws @stroughtonsmith And the Apple Watch to stop having marking as read in messages not buggy as hell. Drives me nutty daily. Thanks watchOS 26!
enthraxxx@stroughtonsmith Is it a competition between Microsoft and Apple Pay to which of them will have created the most unfixable bugs in a month?
gameshack_ 
@stroughtonsmith in a sense it's the (natural) evolution of code completion
Ingo@stroughtonsmith @markgurman I might be wrong but I believe Claude offers a more restrained model for companies that don’t want their data being used to train future models.
poswald@stroughtonsmith @markgurman properly deployed, Claude Code could increase the quality of MacOS and core frameworks by a huge amount. Apple has been neglecting basic bug-fixing, consistency, and stability for years. A brand new mac will spew errors into the console log non-stop from the moment it is booted up. Claude Code is *great* at unit test writing, refactoring, and investigating and fixing small trivial bugs.
Jeff@stroughtonsmith @markgurman I guess I'm just not ready to use an Operating System that's so dependent on Ai that the engineers who developed it can't even fix their own problems without relying on said Ai in the first place.
Markus Müller-Simhofer@stroughtonsmith Sounds useful to migrate tons of legacy C/ObjC code to Swift. I suspect there are many unit tests in place for the frameworks so this doesn’t sound too risky.
@markusms unit tests? At Apple? 😅
That might just have been unique to the WebKit folks
Steve Tibbett@stroughtonsmith @markgurman If it's an internally hosted model (as reported) then it doesn't seem so crazy. Claude seems different than what you describe with Codex, more focused. It generally doesn't make changes you didn't ask for, and you can approve every one. That + PR reviews seems reasonable even for Apple.
Kevin Graaf@stroughtonsmith @markgurman I use NetBSD which explicitly bans Counterfeit Intelligence code
David Boyd@stroughtonsmith @markgurman Maybe they could get it to write good developer documentation.