aichannel

Small sample size over 4 months, but feel like this is directionally accurate

This is why I struggle to accept large code changes in cursor — you dont read the code, you certainly dont understand it, and though product velocity increases, it becomes more brittle

https://x.com/itsalexvacca/status/1935343874421178762

let it make PRs and review it carefully

this is what i am doing these days. i let it go wild. there is a lot of slop in the PRs. i add comments and have claude fix it 

yes, it takes more time and energy compared to “vibe coding” but i don’t lose track of the codebase

Is that more effective / efficient than just writing manually with heavy usage of “Ask” / applying code?

If so, do you have any resources or strategies to learn how to wield AI like this?

I might be using this tool like a boomer

i think, it is more effective is a lot of cases. for instance, i can start a task and then go do chores or something. by the time i am back i expect it to make a PR. i review it on github and post some comments and ask it to address it. reviewing a PR also give me more confidence that i am actually understanding what it is doing  

i have also recently started using this practice so we will see. the key is to break your work into smaller tasks. let it keep track in a markdown file using nested todos 

here is a good starting point: https://github.com/snarktank/ai-dev-tasks

i have added my own git workflow on top of these docs and adopted it for claude code

friendly neighborhood shitcaster // working on @pinetree