Many companies have their first semantic layer in Looker, I know some with 250k lines of code in there. It's a pain to maintain on the model side, but also to use on the consumer side. No one likes the tool, but there's also no great alternative. Cube seems to be the best, but many heads of data engineering seem scared of vendor lock in since there doesn't seem to be a commercial case for a semantic layer only. At the same time, everyone now thinks they can just use LLMs to query data, no semantic layer needed. Maybe use some RAG for the business definitions, have your data modeled in Snowflake and have anyone ask a question. In my experience, text to SQL works great for clean, paper style cases with 5 columns that are clearly distinct, but not for real data. It's easy to build a demo but putting things into production - so you can for example run SQL/Python on the same definitions is a different beast. With all the LLM getting the dataset you want in Python for a quick experiment is still a pain -> we're working on this, the schema/contract can just be your Looker code. would love feedback!