Often, in NL2SQL usecase, the SQL agent generated by the agent/tool needs to be validated for syntax, data types, incorrect object names etc.
This needs building a seperate tool for doing the bigquery dry run and validating the query.
Either the user passes the SQL or the SQL generated by the sub-agent/tool can be validated using this in-built tool, saving the effort in writing the tool code, re-usability and making the code compact and clean.