Apple’s Guideline 2.5.2 blocks AI coding apps like Replit and Anything from updating because generated code running inside an app view is never reviewed.
Key Takeaways
Guideline 2.5.2 predates vibe-coding tools; Apple’s fix for Replit was to open generated app previews in Safari, not stop generating code.
Anything’s co-founder submitted four full technical rewrites over three months; Apple removed the app anyway after the final attempt mirrored Apple’s own suggestion.
The core epistemological problem: reviewers inspect a static binary, but adaptive software generated at runtime by a model produces behavior no reviewer ever saw.
ChatGPT on iOS now distributes third-party MCP-server-based apps through its own internal directory, running against the spirit of 2.5.2 without enforcement – creating a direct contradiction on the same device.
Distribution is moving up the stack: binary to capability (MCP) to intent-based composition; platforms that can hold software in motion survive, those that can’t become CD-ROMs.
Hacker News Comment Review
Commenters are split: one camp sees enforcement as logical given rules that always existed; the other flags selective enforcement as the real problem, not the rules themselves.
The selective-enforcement argument is concrete: the same behavior tolerated inside ChatGPT is being used to block Replit, which invites charges of anti-competitive targeting rather than principled policy.
Notable Comments
@AndriyKunitsyn: draws the selective-enforcement point sharply – a law applied only to some is what generates legitimate outrage, not the law itself.
@jimrandomh: notes devtools have never been compatible with App Store rules; existence of Replit on iOS at all was the surprise, not its eventual blocking.
