Apple's App Store Rule Blocks Updates to AI Coding Tools, Raising Fundamental Questions About Reviewing Adaptive Software
Key Takeaways
- ▸Apple's enforcement of Rule 2.5.2 has blocked iOS app updates for AI coding tools like Replit and Vibecode since March, with Replit slipping from top-ranked to third place in developer tools
- ▸The core issue is epistemological: Apple's review process evaluates a static code artifact, but cannot assess software whose behavior is determined at runtime by generative AI models
- ▸Attempts to comply by routing AI-generated previews to external browsers instead of in-app web views still resulted in rejection and app removal, suggesting the conflict cannot be easily resolved
Summary
Apple is enforcing App Store Rule 2.5.2 against AI coding applications, blocking updates to tools like Replit and Vibecode since March 2026. The decades-old rule requires apps to be self-contained and prohibits them from "executing code which introduces or changes features or functionality." Replit's iOS app has been frozen on the same version for months, causing it to slip from first to third place in Apple's free developer tools rankings, while another app called Anything was pulled entirely after submitting four technical rewrites attempting to comply with Apple's feedback.
The enforcement exposes a fundamental mismatch between Apple's review framework and modern generative AI software. Apple's reviewers can inspect the static app binary during review, but cannot evaluate software whose runtime behavior is determined dynamically by AI models responding to unpredictable user prompts. This creates an epistemological gap: the "wrapper" reviewers approve is not the same as the actual software users experience, which includes AI-generated code previews and functionality. The conflict highlights a deeper structural problem—the entire software distribution ecosystem (versioning, release notes, bug tracking, CI/CD pipelines) was built on the assumption that software remains static after release. Adaptive, generative AI systems that continuously evolve at runtime challenge this foundational premise.
- This exposes a broader incompatibility between legacy software distribution infrastructure and adaptive AI systems that continuously evolve and personalize at runtime
