Working with Claude Code

I develop alone. Planning, design, frontend, backend, deployment. All of it, by myself. So I needed help.

I started using Claude Code a few months ago. I was skeptical at first, but now it’d be hard to go without.

Where I Use It

Refactoring is where I use it the most. “Apply this same pattern to the other files.” That kind of work. Fecit has many screens with similar structures. Browse task record, browse template, browse sub-task. Change one, and the rest need to follow.

I used to open each file and repeat the same edits manually. Miss one, and it comes back as a bug later. Now I just say “apply the changes I just made to this file across the rest.” Nothing gets missed, and consistency is maintained.

I also use it for native module work. iOS is Swift, Android is Kotlin, and there’s the JavaScript bridge. You have to touch all three at once, and few people are good at all three. I’m not. Claude Code understands and modifies code across all three platforms simultaneously, which makes a real difference.

Bug tracking, too. “The keyboard doesn’t come up in this specific scenario — figure out why.” It traces through the code and finds the cause. Not always right, but even pointing me in the right direction saves a lot of time.

What’s Good

Speed, especially for repetitive work. Applying the same pattern to 10 files takes 30 minutes by hand. With Claude Code, it’s 5.

It’s good at reading and understanding the entire codebase. “Find everywhere that uses showLinks and clean it up.” It traces imports, removes unused code, and does it cleanly.

When you develop alone, there’s no one to review your code. Claude Code fills that role to some degree. It catches edge cases I missed, or spots unnecessary code.

What’s Not

Sometimes it does too much. I ask for a small fix, and it tries to refactor the surrounding code. I have to say “don’t touch that part.”

It loses context sometimes. When conversations get long, it occasionally forgets earlier work and repeats the same mistakes. But once you point it out, it remembers and doesn’t do it again.

It can’t test on real devices. Problems that only appear on physical devices but not simulators — I still have to catch those myself. Obviously. But still, sometimes I wish.

It Feels Like Pair Programming

The best way to describe it is pair programming with AI. I set the direction, Claude Code executes. When I say “that’s not quite right,” it fixes it immediately.

The best part is that it’s always available. An idea at 3 AM, an urgent bug on a weekend — it’s always there. For a solo developer, that’s a significant difference.

As of this writing, the Fecit app has about 1,350 commits. Around 630 of those were co-authored with Claude Code. Nearly half. That says a lot about how much I use it, and how much I rely on it.

It’s not a perfect tool. But being able to build and maintain an app of this scale by myself — honestly, this tool is a big part of why that’s possible.

---

A good tool doesn’t replace people. It expands what people can do.