A Month With Claude Code: It's Really Not the Same Thing as ChatGPT

My friend is an independent developer. He takes freelance gigs and works on his own side projects.
Last month he switched his daily workflow over to Claude Code. I sat next to him a few times and watched. Let me share what I actually saw.
First, Let's Be Clear: It's Not ChatGPT
A lot of people assume Claude Code is just "ChatGPT in your terminal." That's not quite right.
With ChatGPT, you copy code in, it gives you suggestions, you copy the result back out, and you manually edit your files. That process gets annoying fast. Change three or five files and you're ready to throw your keyboard.
Claude Code runs directly in your terminal. It can read your files, edit your files, run tests, check git diffs, and even help you commit.
My friend told me that the first time he watched it run npm test on its own and then fix the code based on the error output, he just sat there for a second. This isn't chatting. This is actually doing the work.
Pricing
Claude Code Pro starts at $20/month, that's the subscription tier. If you go with API pay-as-you-go, the cost scales with usage. He hasn't tracked his exact spend, but he says it's "way cheaper than hiring a junior engineer."
What My Friend Actually Built
He took on a small project, building the backend for a booking system for a local restaurant brand. Tech stack was Node.js with PostgreSQL, medium complexity.
His approach: have Claude Code scaffold the project first, define the database schema and basic API routes. Then have it write test cases, run them, and iterate based on failures.
He said getting the backend from zero to a working basic flow took about two days. Writing it by hand the old way, he estimated four or five days.
One detail stuck with me. He asked Claude Code to refactor a messy middleware file. It didn't just clean up the code, it also flagged a potential SQL injection risk he hadn't noticed himself.
The Downsides Are Real Too
First problem: it can only work on the current repository, not across projects.
Sometimes he needs to reference a utility function written in a different project, and he has to manually copy it over. Claude Code can't reach across on its own.
Second problem: limited context.
Once a project has a lot of files, it sometimes "forgets" logic it changed earlier, leading to repeated work or conflicting edits. He's gotten into the habit of having it re-scan the current state periodically.
Third problem, and the biggest one: it doesn't handle complex monorepos well.
Another project in his studio is a large monorepo with a dozen or so interdependent packages. He tried using Claude Code there, and the results were poor. It kept editing the wrong place, or missing cross-package dependencies. For that kind of work, he ended up going back to handling things manually.
Remote server work isn't great either. When he needs to SSH into a server to debug a production issue, Claude Code doesn't help much, since it's mainly built around working with local repositories.
Versus Cursor
They've both used Cursor. The difference is pretty clear.
Cursor is an IDE. You watch the cursor move in real time, code changing line by line in front of you. Good for detail work you need to watch closely.
Claude Code is a CLI. You hand it a task, it runs on its own, and gives you results when it's done. Good for the "let me grab coffee and come back to see how it went" kind of work.
His current habit: small fixes and tweaks go through Cursor, because the feedback is immediate. Larger feature builds, or repetitive refactoring tasks, get handed off to Claude Code.
I haven't used Cursor's team collaboration features myself, but I've heard the team plan allows shared configs. Neither of us has tested that in depth.
Who It's For
If you spend a lot of time writing backend logic, running tests, or doing refactors, Claude Code can save real time.
If you mostly work in a large monorepo, or frequently debug production servers remotely, it might not help much.
My friend's takeaway: it changed the rhythm of how he writes code, but it hasn't fully replaced manual work.
I maintain an AI tool directory at saas.pet (https://saas.pet), updated automatically every day.
Next up, I'll write an in-depth look at Cursor, and whether it actually holds up on large projects.