Vibe Coding: Building an iOS Calorie‑Tracker App with ChatGPT in 400 Lines

Vibe Coding (氛围编程) has become a buzzword in Silicon Valley after Andrej Karpathy demonstrated that, with no prior Swift experience, he could create a complete iOS calorie‑tracking app in just 400 lines of code by conversing with ChatGPT.

Karpathy’s workflow consisted of four dialogue phases: initializing the project in Xcode, adding core functionality, persisting data with AppStorage , and finally deploying the app to his iPhone. ChatGPT acted as a step‑by‑step tutor, providing UI layout code, timer logic, dark‑mode support, and persistence instructions.

The community quickly followed suit, producing games, web pages, and other utilities through “Vibe Programming.” Notable examples include a 10 k‑line Claude‑generated game, a puzzle game built with Claude 3.7 + o1 Pro, and various UI experiments that required only a few hundred lines of AI‑generated code.

Experts stress that Vibe Coding is distinct from merely using LLMs to assist coding. It implies a hands‑off approach where the developer trusts the model to generate most of the code without reviewing each diff, only intervening when errors arise.

While this method can dramatically accelerate prototyping and lower the barrier for non‑programmers, professional engineers still need to audit, test, and ensure maintainability of AI‑generated code. The practice therefore highlights both the potential of AI to democratize software creation and the importance of disciplined engineering practices.