It’s a good observation. Counterpoint - the organization labor is drudge work, maybe with cheap AI labor you can have your KB kept in great shape, and likely this makes it easier for RAG to locate the correct document.
Maybe there is room for both, though I am in agreement that it seems dubious a human should be curating.
I mean that's basically what all these image search indexes do. They look at an image and add a bunch of tags for whatever they see in the photo that can be queried later. It works really well and armed with one of these models and word2vec you can build your own search quite easily.
tensorflow object detection is one such means. Hosted solutions would be Amazon Rekognition.