My Experience Using Claude Code as an iOS Engineer
2
Are you using Claude Code yet?
Introduction
We are living in a flood of AI tools. Just as Cursor became an essential part of the development process, Claude Code is now gunning for its spot. Articles titled "Goodbye Cursor, Hello Claude Code" are everywhere, and an increasing number of engineers are jumping on board. Anthropic actually released the tool they were using internally. Now, Claude Code is quickly becoming a must-have in many engineers' workflows. According to Boris Cherny, the creator of Claude Code, Claude Code itself is now being written—and continuously rewritten—by Claude Code.
Even within the DelightRoom engineering team, our productivity gurus and early adopters were already achieving massive efficiency gains using Claude Code. Honestly, I hesitated to try it at first because of the interface—it's a terminal tool, not an IDE. I was also a bit put off by the pricing. But after actually using it, one thing is clear: Claude Code is definitely better than Cursor. I might be a little late to the party, but here is my experience using Claude Code as an iOS engineer.
What is Claude Code?
Claude Code is a command-line tool for agentic coding. Agentic coding is a development methodology where an AI agent autonomously writes, modifies, and tests code. Anthropic released this tool after their engineers and researchers used it internally for their own coding workflows.
The original creator is Boris Cherny. Fun fact: Boris's grandfather was a computer programmer in the former Soviet Union. It looks like Boris is currently working at Anysphere, the company behind Cursor.
Getting Started with Claude Code
It's super simple. You just install it via npm and type claude in the project directory where you want to use it. Since it's a CLI tool, it might feel a bit unfamiliar at first.
npm install -g @anthropic-ai/claude-codeOnce the installation is complete, just navigate to your project in the terminal and run the claude command.
CLAUDE.md
Running the /init command analyzes your current project and generates a CLAUDE.md file. After this initial setup, every time you run claude, it adds that CLAUDE.md file to its context. If you define commands for running the project initially, frequently used utility functions, coding style conventions, and testing approaches, Claude will appropriately leverage them when needed. Writing this manual essentially gives Claude a reference guide for its autonomous coding. There's no strict format—just write it clearly so it's easily human-readable. You can simply tweak the file generated by the /init command to fit your intentions.
It's also a great idea to use Claude Code when improving the CLAUDE.md file itself. If you want to use settings without sharing them with your team, you can just add a CLAUDE.local.md file to your .gitignore.
Claude Code's SwiftUI Coding Skills
You get a level of precision that you just don't feel with Cursor. It does a great job of selecting and using the latest available APIs while strictly adhering to the minimum iOS version specified in the CLAUDE.md file. When you run plan mode, it understands exactly which files need to be modified. You can toggle between plan mode and auto-accept mode using Shift+Tab. I always use auto-accept mode for simple tweaks, and plan mode for tasks that require a bit more thought.
Claude Code vs. Cursor: A Practical Comparison
We had a minor bug on the alarm edit screen. For video alarms, the sound cell wouldn't disable even if the volume was set to 0. It's a very simple issue that can be fixed quickly once you find the relevant code. An engineer looking at this code for the first time would probably search for the screen's name first. They might search for "Edit Alarm", "AlarmEditor", and so on, or they could search for the text "Play with video sound" to locate the code. I just fed Claude Code the image above along with the bit of information I already knew.
The actual prompt I used was:
> [Image #1] Fix the issue based on this image. diable title label if volumn is 0. near 사운드 in image.Surprisingly, it resolved the issue in the exact same way I would have.
I tried multiple times in Cursor using the exact same Opus 4 Max model, but I couldn't get the desired results.
My Thoughts After Using It
When using Cursor, I still felt like I was the one doing the development. Claude Code, on the other hand, felt like handing the wheel over to Tesla's Autopilot and only choosing the destination. It wasn't quite vibe coding, but I really felt that the engineer's role in the development process has been reduced to choosing which problem to solve and verifying the results. Even when it plans in the wrong direction or misses the point, if you just provide the right context and point it the right way, Claude Code solves the problem almost perfectly.
Lastly, it was especially helpful when navigating the codebase for the first time. Asking Claude Code questions yielded clear explanations of the core logic, which drastically cut down the time needed to understand the code.
I hesitated at first because of the terminal-based development environment, but Claude Code is definitely a step above Cursor. If you're still on the fence, I highly recommend giving it a try. I think it provides more than enough value.
References