Salesforce supports the retrieval of column metadata using SOQL. This metadata includes the column's description. Once retrieved, these descriptions could then be synced to a Snowflake destination (and perhaps other types) in it's field metadata's comment attribute. Snowflake will then display these comments within its UI. Furthermore, the Snowflake comments can be pulled in using SQL, and used to populate dbt YAML files so the documentation is available for dbt-based documentation, as well.

Here's code to pull field-level descriptions from Salesforce:

SELECT QualifiedApiName, Description
FROM FieldDefinition
WHERE EntityDefinition.QualifiedApiName = '<SF_ObjectName>'

Here's the Snowflake SQL for inserting a comment of a column:

COMMENT ON COLUMN my_table.my_column IS 'now comment4';

I have played around w/ doing this in Python, but doing so requires reverse engineering Fivetran's Salesforce to Snowflake field naming algorithm and then setting up an environment where the Salesforce query and Snowflake update code could be synchronized and invoked on-demand and/or at regular intervals. But then it dawned on me that this basically describes what Fivetran's already doing!

So, how about it? I think that it could be pretty seamlessly integrated into and work just like what Fivetran does when syncing data from Salesforce. Only sync the description if it's changed, allow the first description sync to be MARless and include it in a Historical re-sync.