A lot of people I know rate Claude highly.
That is enough to keep me curious. Enough people whose judgment I respect keep talking about it that I keep opening Claude Code again.
I already prefer Cursor for everyday work. That preference has not stopped me exploring. If Cursor disappeared tomorrow, Claude would probably be my next stop. Sorry, GitHub Copilot. That is a joke about my current habits, not a finished comparison of all three.
What impressed me in the trial sessions
In separate sessions, Claude Code handled multi-file work in a way I found genuinely useful. It explored relevant code, planned changes, edited across files, ran tests and iterated when something failed.
What I valued was the sense of delegated work. I could give it a task and let it move through the steps instead of supervising each tiny edit. My subjective impression from those sessions is that Claude Code felt particularly strong on larger delegated jobs.
That is an impression from my trials. Passing tests in a loop is useful. It is not the same thing as proving correctness or security.
Claude Code also has permission modes that change how much it asks before editing or running commands. Official documentation describes an Auto mode that runs actions with background safety checks when that mode is available. The verification loop, in whatever permission shape I was using, felt like part of why the sessions were useful.
Why I still open Cursor the next morning
The main reason is not a clever feature shootout. It is continuity.
My custom skills, hooks, automated checks, project context and familiar editor environment already live in Cursor. I know how to operate that day. I may ask the Cursor agent to inspect or check work as part of that familiar process. My useful role there is still directing the work and reviewing the product, not pretending I win by hand-editing every line.
I enjoy the familiar IDE. Claude Code impressed me as a delegated worker; Cursor currently provides continuity across the working day.
That does not establish a permanent limitation in Claude Code. It establishes where my setup already is.
The surrounding workflow already lives in Cursor
I plan work, bring relevant context into the project, move through stages, use custom automation for checks and then review the result on a testing site. The longer version of that process is in the Cursor preference piece linked above. Those habits are working practices, not exclusive Cursor capabilities. Another tool could host a similar surrounding process. I have not yet rebuilt that surrounding process for Claude Code.
The terminal picture was incomplete
My earlier mental picture of Claude Code was strongly terminal-shaped. I preferred having the work sit together in a familiar IDE, so that association mattered more than it should have.
The current product picture is wider than that. Anthropic’s Claude Code overview (opens in a new tab) describes the same agentic coding tool across the terminal, IDE extensions, a desktop app and the web, including VS Code and JetBrains integrations. I have not personally tested every surface. The factual correction still matters: Claude Code is not limited to terminal use.
Ordinary local project access is also easy to confuse with a different Claude setup. Claude Code is built to read and edit an authorised local project. Connecting external tools through MCP (opens in a new tab) is a separate idea. Configuring filesystem access for Claude Desktop through local connectors or extensions is another setup again. Cursor also uses MCP when it needs external tools. None of that means Claude Code requires a separately configured filesystem MCP server just to work in a local repo, and none of it means Cursor never uses MCP.
The subscription joke that is not really a joke
I have previously hit Claude usage restrictions that interrupted the way I like to work. I am still unsure whether that was a daily limit, a time window or something else. I wrote about cancelling that short Claude subscription for reasons of fit, not because the product was useless.
I already spend $200 a month on my main Cursor subscription. The joke underneath is simple: I do not want to pay for two expensive overlapping coding setups just to keep both workflows warm.
The next experiment has not happened yet
The fairer test is the one I have not run.
I plan to adapt my Cursor skills and hooks for Claude Code in a later trial, then assess the full development workflow rather than a few impressive sessions. Formats, events, permissions and integrations may need real adaptation.
The open question is the only conclusion I trust for now:
If Claude Code had an equally familiar and complete surrounding setup, would I still return to Cursor?
I do not know yet. That is why I keep trying.
