>This also removes the typical more organic selection process of libraries and tooling, replacing it with whatever was most prevalent in the LLM’s training data

Another article written by someone who doesn't actually use AI. Claude will literally search "XYZ library 2025" to find libraries. That is essentially equivalent to how it's always worked. It's not just what is in the dataset.

> "XYZ library 2025"

I'm fairly sure you made a typo, but considering the context, it's a pretty funny typo and would kind of demonstrate the point parent was trying to make :)

I agree with you overall though, the CLI agents of today don't really suffer from that issue, how good the model is at using tools and understanding what they're doing is much more important than what specific APIs they remember from the training data.