Everything about vibe coding 🤖
Tools, tips, tricks, workflows, best practices, you name it.
Let the vibes guide you

Vibe Coding

Do you still write most of your code?

The ratio I write code is sometime 20/80 and sometime 80/20 depending on the requirements of the feature.
And Yes Opus is really helpful in my complex tasks compared to other intelligent models, yes it is expensive but It does mostly things in one go.
It is just my opinion with the Opus, other models also do well like GPT-5.2-Codex-Max is cheap and also performs like Opus sometimes.

When I went back to the feed from this post, my vote was no longer visibly applied. Tapping a poll option gave a “something went wrong” pop up on the bottom (Android app)
Thought you’d wanna know if it wasn’t on your radar already : )
I’ve been pretty impressed by plan mode with Claude code for home messing around type projects. Really like that it will ask questions to increase the odds of getting it right or something on the right track. Tried it in a larger C++ codebase and I’ve found it could use explicit reminders on things like helper/utility functions in that context, else it would sometimes make new ones. Solid overall for sure.
I think it’s helped me pick up on some things a little bit quicker, but hard to say

Hard to tell. It really depends on the type of work I’m doing. I feel like I’m still in an adjustment phase and my workflows change every week (which is exciting, actually)
There are problems on which I like to think by code, but ale the boring standard stuff, I don’t see a point of coding lots of things myself.
That said, I’m not a hardcore AI user, my job isn’t about speed nowadays.
Mostly using Claude Code and it serves me well.

I do write most of my code, but I am not going to say that I don’t use AI to simplify my life.
<ol>
<li>AI might give me the skeleton and then I modify everything, so then it’s still my code</li>
<li>It’s useful for writing unit tests. It is funny when they fail, only for me to realize that its unit test is valid and my code has potential bugs.</li>
<li>AI can do some amazing code reviews.</li>
</ol>
<s>What I don’t do with AI is tell it: This is not working, please fix it. Even with Github Copilot and Claude Haiku 4.5 (which is supposedly better than Claude Opus 4.5), it is not able to follow through and solve the problem I described.</s>
What I don’t do with AI is tell it: This is not working, please fix it. Even with Github Copilot using Claude Opus 4.5, it is not able to follow through and solve the problem I described.
For those that read this and get confused. GitHub Copilot is not an LLM. It’s a tool that uses an LLM I choose.
<img src="https://media.daily.dev/image/upload/s--ucExAn0L--/f_auto/v1767714612/ugc/content_14a3a7cf-b6f2-4042-b2e5-525e91930614?_a=BAMAK+ZW0" alt="image">
I will continue to debug, one line of code at a time.