I think lots of issues with the integration of CC or other TUI with graphical IDEs will be solved with stuff like the Agentic Coding Protocol that the guys at Zed at working on https://www.npmjs.com/package/@zed-industries/agentic-coding...

I trust zed to get it right over cursor with their continual enshittification.

Claude Code is totally different paradigm. You don't edit your files directly so there is no tab autocomplete. It's a chat session.

There are IDE integrations where you can run it in a terminal session while perusing the files through your IDE, but it's not powering any autocomplete there AFAIK.

are people viewing file diffs in the terminal? surely people aren't just vibing code changes in

Yes. I manually read the diff of every proposed change and manually accept or deny.

I love CC, but letting it auto-write changes is, at best, a waste of time trying to find the bugs after they start compounding.

it seems like CC is king at the moment from what I read.

I currently have a Copilot subscription that has 4.1 for free but Sonnet 4 and Gemini Pro 2.5 with monthly limits. Thinking to switch to CC

I am curious to know which Claude Code subscription most people are using... ?

Yes or running claude code in the cursor/vscode terminal and watching the files change and then reviewing in IDE. I often like to be able to see an entire file when reviewing a diff, rather than just the lines that changed. Plus it's nice to have go-to-definition when reviewing.

Yes, it shows you the file diff. But generally, the workflow is that you git commit a checkpoint, then let it make all the changes it wants freely, then in your IDE, review what has changed since previous commit, iterate the prompts/make your own adjustments to the code, and when you like it, git commit.

Depending on what I'm doing with it I have 3 modes:

Trivial/easy stuff - let it make a PR at the end and review in GitHub. It rarely gets this stuff wrong IME or does anything stupid.

Moderately complex stuff - let it code away, review/test it in my IDE and make any changes myself and tell claude what I've changed (and get it to do a quick review of my code)

Complex stuff - watch it like a hawk as it is thinking and interrupt it constantly asking questions/telling it what to do, then review in my IDE.

I review and modify changes in Zed or Emacs.

Apparently they are, which is crazy to me. Zed agent mode shows modified hunks and you can accept/reject them individually. I can't imagine doing it all through the CLI, it seems extremely primitive.

I use windsurf to check the diff from Claude Code.

If there’s a conflict you just back out your change and do it again.

I just accept all and review in my editor.

that's what lazygit in another terminal tab is for