I’ve been using AI coding assistants for a while now. I started with Trae, specifically its SOLO mode. It felt like the WebStorm of AI editors — everything worked out of the box, and I never had to think about how to talk to it. It just understood me.
Then I switched to Claude Code. And honestly? It’s been a wake-up call.
Don’t get me wrong — Claude Code is powerful. But after months of using Trae’s SOLO mode, I’ve gotten lazy. Claude Code doesn’t hold your hand the same way, and I found myself frustrated by things that I used to take for granted.
Here’s what I ran into, and how I eventually fixed (most of) it.
The Problem: Too Many Manual Steps
The first thing that bugged me was the workflow. With Trae’s SOLO mode, I’d type a request, and it would automatically break things down: first it analyzes the problem, then it comes up with a plan, then it writes the code, and finally it tests everything. I just sat back and reviewed the result.
Claude Code doesn’t do that by default. It’s more of a conversational partner — you ask, it answers, and you keep going back and forth. That’s fine for small tweaks, but for bigger tasks, I kept wishing for a single command that would trigger the whole analysis → plan → code → test cycle.
The fix: I installed the superpowers plugin globally. It gives Claude Code exactly that kind of structured workflow. There are two ways to use it:
- Auto mode: Claude decides on its own when to use the structured approach based on your request.
- Manual mode: You type
/brainstormingbefore your prompt to kick things off.
I’ve been using the manual trigger, and it works. But I still forget to type it sometimes. Old habits die hard.
The Confirmation Overload
This one drove me nuts. Every time Claude Code wanted to create a new file or modify existing code, it would ask for my permission. One by one. For every single change.
I get it — it’s safe. But when you’re refactoring a small module and there are 20 files to touch, approving each one feels like babysitting. Trae never did that. It just ran with it and only stopped when something genuinely risky came up.
The fix: I found out about the --enable-auto-mode flag. You launch Claude Code with claude --enable-auto-mode, and it stops asking for every little thing. High-risk operations still trigger a confirmation, but routine file edits happen automatically.
The bigger question was whether I could make this the default, so I wouldn’t have to type the flag every time. Turns out you can. I asked Claude to add this to my global settings:
json
{
"permissions.defaultMode": "auto"
}
I put that in ~/.claude/setting.json, and now auto-mode is always on. It also works alongside superpowers — the structured planning kicks in when needed, and the auto-approval handles the execution. That combination finally feels right.
Forgetting to Document
After a long debugging session, the last thing I want to do is write down what I just did. But I also know I’ll need it later — either for myself or for someone else on the team.
Trae didn’t have a good solution for this, so I was used to just… not documenting. But Claude Code can do better.
The fix: I installed the everything-claude-code plugin. It adds a prompt after every conversation, asking whether I want to save the problem and solution as a note. It also lets me choose between saving it globally (for all projects) or just for the current project.
Now I still forget sometimes, but at least the system nudges me. That’s more than I had before.
Rolling Back Changes Without Git
Here’s something Trae got right: if you mess up, you can undo all changes from the current session with one click. And it doesn’t depend on Git commits. You don’t have to remember to commit before starting, and you don’t end up with a messy branch history.
I assumed Claude Code would lack this, but I was wrong.
The fix: Claude Code has a built-in checkpoint system. Every time it makes changes, it saves a snapshot. To roll back, you just:
- Hit Esc twice (my favorite — it’s fast), or
- Type
/rewindin the chat
That brings up a menu showing all the checkpoints. You pick one, and it reverts everything — no Git commits required. It’s clean, it’s fast, and it works exactly the way I needed it to.
Final Thoughts: Unlearning and Relearning
I spent a few weeks tweaking Claude Code to behave more like Trae. And to be fair, most of the problems turned out to be solvable — plugins, settings, a few config tweaks. But the fact that I had to do all that research and setup is exactly the point.
Trae’s SOLO mode spoiled me. It gave me a polished, opinionated workflow that I didn’t have to think about. Claude Code is more like a toolbox — powerful, flexible, but you have to build your own workflow on top of it.
I don’t think one is better than the other. They’re just different tools for different mindsets. If you want something that works out of the box and you’re okay with fewer knobs to tweak, Trae is great. If you want full control and don’t mind tinkering, Claude Code can get you there too — once you know which plugins to install and which config flags to set.
Now that I’ve got it all dialed in? I’m happy. But I’ll never forget how nice it was to not have to think about any of this in the first place.
