Yep, we've built some really useful agents with some simple tools (3-5 templated snowflake queries with really good descriptions). The LLM is useful for shaping your question into function params and then interpreting the results based on the context it got from the tool description.