I was never confident enough to go into yolo mode in Claude Code after I had such a bad experience with Cursor…or I failed quite hard with the prompts. 🤷
Last weekend, I wanted to try it out for a task where I didn't want to sit next to Claude, watch it, and prevent it from failing. So I’ve made a new attempt. “We” planned the task together; I tried to be as precise as I could be with the prompt and got a first draft as a result. Then we iterated and refined the plan until I was confident with it.
On the weekend, we had planned some family time. It was the best opportunity to let Claude go wild with the planned task. Later in the evening when I was back with the family, I was a little excited about how it worked out. I took a first look, and Claude had completed the task. The result was working, and the tests were green. So this looked promising. I peeked at some files, and it looked good.
After I did a full review on the next day, I was quite happy with the result. There were some small adjustments where Claude was not following the rules exactly, but this is ok. Like checking a PR from another colleague.
Maybe I should try this more. One idea would be that after a finished workday, I go into yolo mode with Claude and review the results on the next day. But before I can do this with confidence, I need to refine my Claude settings. The settings are there, but I never touched them, really. There is potential here. :)
Today I had an interesting AHA moment. I never thought about my meeting participation when I'm a part of a team. But I automatically participate more in smaller teams and less in bigger ones.
Why is that so?
A simple explanation could be that in bigger teams, there are often people who are quite dominant, talk a lot, and usually know more than me. Maybe this influences me subconsciously. I’m not a dominant or submissive person. So I typically withdraw then and let the others be dominant. And say something if I have to or need to.
Perhaps it also has something to do with me not really being a speaker. I typically can talk a lot in a space where I’m confident. Most in small groups or 1-to-1. One example is presentations in school. I hated them. Preparing them was easy. But actually standing up in front of more than 5 people made me nervous. And I hated that feeling. Not the good part of feeling nervous.
What can I do with this new insight? I don't know. 🤷 I have to think about it, and I'm happy that I now have something I can work with. For the future, it would be best to get rid of this limiting behaviour.
Kicked off the migration to react-twc in one of my projects.
For simple components, it works well. But if they are not just styled, for example, if they need to render a different element, it gets verbose. <MyComponent as=“section” /> now needs the asChild prop and then the section element as a child. It’s not that bad, but I think it’s ugly.
As someone who also uses many Radix components, I’m familiar with it. So it is ok, but it still looks ugly.
For variants of the same element, I migrated to react-twc with cva. Which gives a clean look and makes clear what needs to be used.
And for the rest, for example, labels and inputs, I kept them as is, for now. The label and inputs are wrapped with a div. Also, I need to pass down the Radix Form elements with the as prop.
Thoughts
For some components, I have to refactor the code. At the current state, I don't know whether it's worth using react-twc or not. For simple components, I think it is awesome. Furthermore, for components with plenty of variants, by using CVA. But for a complex one, I have to think about it and work out something that fits.
I’ve found a way I can personally handle tasks and tickets at work.
The problem is that most of the time, no matter how much I’m into a task, I forget plenty of things. I don’t know why. So I came up with a simple structure and Obsidian.md as a tool. On how I can express my thoughts on such topics and also save them for later.
When writing with cursor, it is like talking with your silly self and thinking the outcome is really good.
Writing with CC is like writing with an elaborate sane person, who really knows more than you. (Most of the time.)
When writing a prompt to either of them, I have this mindset above for each in my head, and this is how my prompts look. Maybe this is the reason why my outcome on cursor is not as good as in cc?
Nevertheless is gain something from both of them, so the current state is ok for me.
This is one of my projects that has a low priority and doesn’t get the love it deserves. But in the end, this is one what saves me a lot of work and I need to adjust only a JSON entry after I’m done with a client.
There is a post about how I migrated to WEZterm, which I still need to write at some time in the future.
Preliminary skirmish
But let’s assume I already wrote the post. I’m really happy with the terminal. I needed to migrate away from Alacritty+Tmux because of the lack of Windows support from Tmux. Overall it was a good decision to migrate. The only thing that currently bothers me is that there is a “bug” where the font rendering misses some characters but brings them back on config reload. (cmd+r)
Here starts the main post
After starting my journey with Claude Code, I needed a handy shortcut to open CC in a 1/3 split pane. Here is the part of my config, which is doing the Job:
Over the weekend, I've tried to work on a conversion from HTML to PDF and to DOCX.
It all started with wkhtmltopdf. I noticed (much too late) that this project is not maintained anymore. So a replacement was needed. I found one with Puppeteer, and it works really well. It feels a bit too much for the task, but it creates a PDF out of my static HTML page, which is enough.
While at it, I fired up Claude Code and asked to do the same for the HTML to DOCX conversion. Oh boy, I didn't know that this task was a nightmare. For years I had this conversion planned for a tool but never found the time to do it. Now with CC, I thought it would be an easy task that is done in minutes.
After some more than an hour, I stopped and thought, this is not worth the effort. I don't blame CC or the tools I've used for this job. I blame the DOCX format for being so shitty.
I've tried pandoc, html2docx and html-2-docx, and none of them produced the output I've expected or needed. There are also commercial tools, which I thought would be too much for what I require. And it still is not worth paying money for something I require so rarely I could do the adjustments myself every time I require it.
So I stepped back and thought, maybe this feature needs to keep lying in my to-do list for the next years. I had already put enough effort into it, and it will not save the time I already invested in it.
Since March, I have used Cursor on and off for plenty of tasks. For example, I did a complete Redux to Redux-Toolkit migration for one project, in which I learned a lot. Mostly, for which tasks I should use Cursor and which not. If you start with Cursor, you should use it to implement a bigger feature, to learn all the quirks. It will take time, and I value the time using it for the Redux→RTK migration. And now, I typically use Cursor for smaller tasks. Or just break down the bigger ones and enhance them so they are suited for Cursor.
For a while, I struggled with the right model. The auto mode worked in some way, but I was not perfect. So I tried to explore the models directly and wanted to learn, which I can use for which task. But since Claude released Sonnet 4, I exclusively use this model for everything. It is astonishing how well Sonnet 4 works.
Recently, I have found a good workflow. I collect some Tasks and execute them one by one. And while Cursor/Sonnet 4 is thinking and working, I do some chores in the House. Which is a Win-win for all of us here. 😅
I trust Sonnet 4 so much, that I'm thinking of trying out Claude Code. Which, in theory, fits better into my terminal workflow. Cursor is a fork of VSCode which is too much UI distraction for me. But I appreciate the diff tool, which works good for me. And I never found a good one for the terminal.