Replit shipped its first iPhone update in four months as of mid-May, and the delay had nothing to do with the product. The Replit iPhone app update is live after an App Store review issue blocked releases since March, holding back Agent 4, parallel agent support, team collaboration via merge flows, and cross-workspace project viewing from every iPhone user. Agent 4 was unveiled in March; the features are only reaching iOS now.
The holdup was a policy collision. Apple reportedly began pushing back on new Replit versions in March over how AI-built apps were previewed on-device. CEO Amjad Masad confirmed the resolution publicly today: "We worked things out with Apple, and just published our app for the first time in four months."
The stakes extend well beyond one app. AI-assisted app development contributed to an 84% jump in App Store submissions in a single quarter, per The Information. Apple's decisions about what these apps can do on iPhone are shaping the architecture of an entire software category at the moment it's growing fastest.
Why the Replit app was delayed by Apple
Apple's Guideline 2.5.2 requires App Store apps to be self-contained. They cannot download, install, or execute code that alters their behavior or introduces new functionality after passing review. The rule predates AI coding tools entirely. Vibe-coding apps that execute AI-generated code at runtime land in that prohibition not because of a new policy, but because of what they do at a technical level.
Apple clarified that it has no rules specifically targeting vibe-coding as a category, and said it maintained consistent communication with affected developers, including at least three phone calls over two months to explain what compliance would require.
Developer Program License clause 3.3.1(B) adds a second constraint: downloaded interpreted code cannot change an app's primary purpose. Together, the two rules mean Apple's reviewers can't fully vet an app at submission if that app generates and runs its own executable code during use, which is the deeper reason enforcement happened, not just the literal text.
The contrast with Xcode is worth noting. Apple added support for OpenAI and Anthropic agentic coding tools to Xcode, letting developers autonomously write, build, and run code inside Apple's own IDE. The distinction Apple appears to be drawing is between a professional developer environment producing verifiable software and a consumer app executing generated code inside the iOS runtime. Whether that distinction is the right one to draw is a separate question; that it's Apple's to make is not.
What recent approvals suggest about complying with the rule
Among the vibe-coding apps that have cleared review, a pattern has emerged. The iOS app handles prompting, editing, collaboration, and project management. What it no longer does, in cases seen so far, is run the generated code inside the native host. Those previews move to a web browser instead, keeping the iOS container static and reviewable, TechCrunch reported. This is a pattern seen in recent approvals, not confirmed Apple doctrine, and notably not what Replit has disclosed about its own fix.
Lovable appears to have avoided the friction seen elsewhere by designing around the constraint from the start. Its app launched on iOS without the delays Replit encountered, positioned around creating "working websites or web apps," language that routes execution to the browser by design rather than retrofitting compliance after a rejection.
Based on the case history available, generating, editing, and managing projects inside a native iOS app appears permissible; executing the generated code inside that same iOS runtime appears to be what triggers review problems. That's an inferred boundary from reported cases, not explicit Apple guidance, but developers entering the category now have enough precedent to build around it before submission.
Uneven enforcement made the rule harder to interpret than the rule itself
Apple applied Guideline 2.5.2 to Replit, Vibecode, and Anything starting in March, but the outcomes varied considerably. Replit spent four months in negotiation. Anything was removed, restored, then pulled a second time after Apple told the company it couldn't market itself as an app maker, a complaint that extended beyond technical code execution into product positioning, TechCrunch reported last month. That second removal signaled the line Apple was drawing wasn't fully contained in 2.5.2 alone.
Anything's case illustrates what navigating this looked like in practice. After Apple rejected co-founder Dhruv Amin's proposed browser-preview fix and removed the app late in March, the team rebuilt its core experience inside iMessage and posted publicly that "Apple is scared of vibe coding." Two days later, the app was reinstated, per aiHola.
What secured that reinstatement isn't clear from public reporting, and whether the public pressure played any role remains uncertain. Developers navigated this through iteration, appeals, and possibly public pressure, not a published compliance standard.
What Replit changed architecturally to satisfy review also remains undisclosed. The original dispute centered on how AI-generated previews were handled on-device, but whether the fix mirrors the browser-based pattern seen elsewhere isn't confirmed. As of last month, Vibecode remained blocked from releasing updates. Replit's resolution is one company's outcome, not a signal that enforcement is finished.
What's still unresolved
Replit's return establishes that vibe-coding apps can survive on iPhone. The cases so far point toward a structural separation between where AI builds and where the built product runs, with the iOS app handling creation and the browser handling execution. Companies that architect for that split from the start are likely to have a cleaner path through review than those who negotiate their way to it after months of blocked updates, though the Anything case shows even that approach isn't a guarantee.
Apple has published no guidance specific to generative app builders. Replit's exact technical fix remains undisclosed. Vibecode is still blocked. The pattern is clearer than it was in March, but what exists is case-by-case enforcement of an existing rule, applied inconsistently across a category that didn't exist when the rule was written. With no formal framework in sight, expect more companies to hit the same wall, and expect Apple to keep drawing the line one review at a time.
Comments
Be the first, drop a comment!