I've been blown away by the quality of work that Anthropic Claude Code produces compared to Cursor, and I want to give you practical tips to get the most out of it.
Tips for Claude Code
1) Press Shift-Tab to get into "Plan" mode. Use that generously to figure out what you're going to have the AI do before it does it.
I had no idea about plan mode! That's a real quality of life improvement. I would always append "Don't implement anything yet, I just want to hear your plan" to all of my strategy messages 🙃
- Haniff Carroll, Frontend Engineer
2) Drag reference images directly into the terminal.
3) Run `/init` to automatically create the CLAUDE markdown file. Ask Claude to "remember" something and it will add it to this file. Put all your rules (think Cursor rules) here.
4) Since there's no restore function, push to your repo often at key checkpoints so you always have a backup.
I also ask it to keep documentation up to date, track development progress and generate a code index. This all helps provide and maintain context over time.
- Gerlando Piro, Web Engineering Consultant
5) If you're using VS Code or JetBrains, use the accompanying extension for better functionality.
6) Use the `/compact` command when you hit milestones to compress down your context. If you don't, it'll force you to do it anyway.
7) Ask it to break down large tasks and run subagents simultaneously for faster completion.
8) Ask Claude to double-check its work, look for edge cases, suggest improvements, and ensure nothing broke.
9) Remember that Claude can research and browse the web for additional context or documentation.
I've been kind of shocked at how much better it is. I love how it writes less (but better) code, breaks the tasks down into subtasks by itself, runs tests by itself, etc.
To get started
Claude Code: https://www.anthropic.com/claude-code
VS Code extension: https://marketplace.visualstudio.com/items?itemName=anthropic.claude-code
FAQs
How do you manage subagents without completely losing your mind? Also alongside the "no restore function" thing — multiple agents all at once feels like a minefield for error...I haven't been able to get past it yet.
I've found I need to be careful with what I give to subagents, it needs to be small tasks that are hard to mess up. It's definitely a minefield, though. Git's the savior here
So is Claude code an app like Cursor? Does it require development skills to setup?
Once you get it set up and get the VS Code extension, you can open it up on the right side of VS Code (or Cursor) just like the chat in Cursor. You do many of the same things with it there, so it functions very similarly to Cursor!
