Working With Claude Code

Claude Code represents a major change for how I write code. I now default to working in CC and periodically opening my editor. I no longer worry about remembering bash syntax or the exact docker/npm/hatch commands I need to start my server. I don't worry about the syntax to run my tests or how long it will take me to debug my suite. I worry about breaking down the task I'm trying to complete into deliverable chunks.

Claude Code is an incredible tool, so far the best "agentic ai" code writing tool I have found. I've tried quite a few of them, the two I've spent the most time in are Cursor and Aider.

Cursor was great but I don't enjoy writing code in VSCode. I'm historically a Vim user and their Vim bindings feel clunky. Right now, Doom Emacs is my Vim of choice and I quite like it as an editor. When I'm writing the code myself I don't want to fight the editor. Cursor (mostly VSCode) always feels like I'm fighting the editor.

Aider is pretty great. I really like the ability to switch models, even though I rarely use anything that's not Claude. It also lets me use my own editor. It doesn't support MCP, which is a big detractor, another blog post is coming about MCP. It also commits for you, which I'm not a fan of. I want to be able to review the code the machine is writing. Doing that requires my own management of git.

The Claude Code configuration is amazing. Giving it a markdown file of how you want it to operate in human readable language is such a time saver. The relatively easy management of MCP servers is great. I can't get over how great it is. I'm going to write another blog post just on MCP.

Claude Code is like pairing but easier, whenever you're ready to go it is too. It's not as good as a human just yet but it saves an incalculable amount of heavy lifting. It's not always right, the code isn't clean but it's a place to start.