プログラミング嫌いのエンジニアがClaudeだけでiOSアプリを完成させた体験記 An engineer who dislikes programming describes how they built and shipped a complete iOS a…

This article is a first-person account from an engineer who openly dislikes programming yet managed to ship a complete iOS app using only Anthropic's Claude, specifically the Claude Code CLI agent. It serves as a useful data point on how far conversational AI has lowered the barrier to solo app development.

The author admits to having little fluency with Swift or Xcode. Rather than hand-writing code, they describe driving Claude Code through natural-language prompts: stating requirements, pasting build errors, and letting the agent propose patches and configuration steps inside the Xcode project. The workflow is iterative — the human acts mostly as a product owner and reviewer, while the AI handles the translation from intent to working Swift implementation. For someone who finds the act of coding itself unpleasant, this arrangement appears to have been the decisive factor in actually finishing and publishing the app.

The broader context matters here. Since 2024, Anthropic has been pushing Claude Code as a terminal-based coding agent, joining a crowded field that includes GitHub Copilot, Cursor, Windsurf, and various Devin-style autonomous agents. The category often labelled as vibe coding — describing what you want rather than typing the code — has become increasingly viable for non-trivial projects. iOS development is a particularly interesting target because Xcode introduces many environment-specific pitfalls (signing, provisioning, simulator quirks, SwiftUI versioning) that have historically frustrated newcomers. An agent that can read an error log and suggest a concrete fix removes a large portion of that friction.

That said, the experience is not entirely friction-free. App Store submission, certificates, provisioning profiles, and review-related decisions still tend to require human judgement, and the article implicitly suggests that some manual steps in Xcode remained necessary. It is also reasonable to expect that code quality, architectural coherence, and long-term maintainability of AI-generated apps may vary, though the author does not dwell on these concerns.

Rather than a deep technical tutorial, the piece reads as a personal report on what is now achievable for a motivated non-specialist. Stories like this are likely to multiply as agentic coding tools mature, and they reinforce a shift already visible across the indie developer community: the bottleneck for shipping small apps is moving from coding skill toward product thinking, taste, and the ability to communicate clearly with an AI collaborator. Whether that shift will hold up for larger, more complex codebases remains an open question, but for a single-developer iOS app it appears, at least in this case, to have worked.