Who is cleaning up all the garbage LLMs generate? | Hacker News Reader

Who is cleaning up all the garbage LLMs generate?

6 pointskbrannigan7 comments

So my one terabyte drive got full and I realized after months of CLI sessions I generated a lot of markdown files in a lot of images, all sort of other documents they all seem sorta kinda useful, and I too afraid to delete them because they might break something

How do you deal with that?

View on HN

Deleting files is easy. Decommissioning production services is not. It's too early to say for me, we've only decommissioned one vibe-coded service at my Day Job and it wasn't any worse than decommissioning any other legacy service has been. But it became legacy much faster - a couple of months.

What does that mean. Taking them down because the project failed or just replacing it?

Replacement. Getting all customer requirements and migrating them to a superior service implementation.

Legitimate question. In my projects, where I do have a ton of files, but kind of suspect not nearly as you have, I do occasionally instruct agent to review old files and decisions and archive them or delete.

Sometimes it is useful to know how we came about some decision or feature, but in general you often don't need them and they would contain prose that would confuse agent. So you remove it.

I don't use Claude Code mostly because it always loved to generate reports and there is no way I will read it's lengthy reports. I try to have documents I can actually read if I want to.

Hope this helps, but some lifecycle of your documents and requirements is useful to have.

Have agents write scratch output outside the repo, while anything that matters lives in version control. That way you can purge the scratch files whenever you like.

I just delete it and go back to the commit if it matters in the future or generally if the code changed a lot just ask the probs newer smarter model to make the file again

Hmm, hence the issue. It’s like chopping down a tree to make a box of matches. I wish the next generations of models don’t rely too much on markdown documentations which seems to be a human way of not writing code. Sometimes the code is just a little bit more structured to be honest , often times I feel like I might as well just write the code.

Reply on news.ycombinator.com