I really don't know what is it, but Claude Code just seems like an extremely well tuned package. You can have the same core models, but the internal prompts matter, how they are looking up extra context matters, how easy is it to add external context matters, how it applies changes matters, how eager is it to actually use an external tool to help you matters. With Claude Code, it just feels right. When I say I want a review, I get a review, when I want code, I get code, when I want just some git housekeeping, I get that.

I haven’t found massive performance between tools that use the same underlying LLM

The benefit of Claude Code is that you can pay a fixed monthly fee and get a lot more than you would with API requests alone.

That has not been my experience, Copilot using Claude is way different than claude code for me. Anecdotal, and "vibes" based, but it'd what I've been experiencing.

I use neovim so claude code makes more sense to me. I think having the code agent independent from the code editor is a plus.

I use vim for most of my development, so I'm always in a terminal anyway. I like my editor setup, and getting the benefits of a coding assistant without having to drastically change my editor has huge value to me.

I went on n a bit of a YouTube frenzy last weekend on getting an overview of agentic tools.

A lot of people are saying that cursor is much worse than Claude Code who have used both.

Having spent a couple of weeks putting both AIDE-centric (Cursor, Windsurf) and CLI-centric (Claude Code, OpenAI Codex, Gemini CLI) options through real-world tasks, Cursor was one of the least effective tools for me. I ultimately settled on Claude Code and am very happy with it.

I realized Claude Code is the abstraction level I want to work in. Cursor et al still stick me way down into the code muck when really I only want to see the code during review. It's an implementation detail that I still have to review because it's makes mistakes, even when guided perfectly, but otherwise I want to think in interfaces, architecture, components. The low level code, don't care. Is it up to spec and conventions, does it work? Good enough for me.

The native tool use is a game changer. When I ask it to debug something it can independently add debug logging to a method, run the tests, collect the output, and code based off that until the tests are fixed.

I put Cursor as 4th of the tools I have tried. Claude Code, Junie, and Copilot all do work that I find much more acceptable.