I use Claude Code every day. Not just for writing code, but for copy, workflows, documentation, debugging logic that has nothing to do with a function. If you're only using it to generate boilerplate and fix bugs, you're getting maybe 30% of what it's capable of.
By Nick Marello
Here are five things that have actually changed how I work.
1. Set up a CLAUDE.md file
Every new Claude Code session starts cold. No memory of yesterday, no idea what stack you're running, no context about what you've already decided and don't want to revisit.
The fix is a CLAUDE.md file in your project root. Claude reads it automatically at the start of every session. I put everything in there that I'd otherwise spend the first ten minutes re-explaining:
- Stack and versions
- Folder structure and naming conventions
- Decisions that are already made
- What I'm currently working on
Mine is probably 200 lines at this point and I update it as the project changes. It's the highest-return thing on this list for the least amount of effort.
2. Ask for a plan before it starts touching files
Before Claude makes any changes, especially across multiple files, I ask it to write out what it's going to do first and wait for me to approve it.
Something like: "Before you make any changes, write me a step-by-step plan. I'll tell you when to proceed."
Claude will sometimes confidently solve the wrong problem. Getting the plan in front of you before it starts costs 30 seconds. Untangling what it did after the fact can cost 30 minutes. I've had sessions where this one habit alone saved a full hour of cleanup.
3. Commit before every task
Claude moves fast and touches multiple files at once. Sometimes it fixes one thing and quietly breaks three others. The answer isn't to slow it down. It's to have an easy out when things go sideways.
I commit before any task that's going to touch more than one file:
git add . && git commit -m "checkpoint before refactor"If Claude goes off the rails, git checkout . and you're back. I do this reflexively now. Four seconds of habit that has bailed me out more times than I want to admit.
4. Build slash commands for recurring tasks
Claude Code lets you create custom slash commands via a .claude/commands/ folder. You write a markdown file, name it, and Claude treats it as a saved prompt you can call with /command-name.
The ones I use most:
/handoffwrites a session handoff document so the next Claude session picks up exactly where I left off, with full context on what was done, what's in progress, and what to do first/cmsaves something to Claude's persistent memory so I never have to repeat myself across sessions/bugreads the relevant code, identifies the root cause, and fixes it with the minimal change required
Writing a command the first time takes maybe five minutes. If you're typing out the same instructions more than twice a week, it should probably be a command.
5. Use it for work that has nothing to do with code
Claude Code is a terminal-based AI agent. It's exceptional at code, but it's also just really good at working with files and context, which means it's useful for a lot of things people don't think to try it for.
I've used it to draft marketing copy, reorganize documentation, build out content structures, and run writing workflows where the project files themselves are the source material. It works well for this stuff because it already has context about what you're building, context that a fresh chat window doesn't have.
If it lives in your project, it's probably fair game.
Where to start
Pick one. The CLAUDE.md file is the easiest entry point and it makes everything else work better once Claude actually knows your project.
This is exactly how I build prmptVAULT.ai and everything at Marello.Productions. Every feature, every fix, every deploy. Claude Code with the right setup is the fastest way I've found to ship real products.
Nick Marello is the founder of Marello Productions and prmptVAULT.ai.
AI Playbook is a running series on what's working in my workflow. Week 2 coming soon.