Consistent with the recent work to allow the API to accept some primitives as well as specific objects, the schema argument to LoadJobConfig* could accept a list of dicts. Currently it requires constructing a bunch of SchemaField objects:

    schema = [
        # could be just column_dicts
        bigquery.schema.SchemaField.from_api_repr(column_dict) for column_dict in column_dicts
    ]

     with open(empty_path, "rb") as source_file: 
        job_config = bigquery.job.LoadJobConfig(
            schema=schema, write_disposition="WRITE_TRUNCATE"
        )

        job = gbq_client.load_table_from_file(
            source_file,
            table_ref,
            location="US",
            job_config=job_config,
        )

We're trying to move to using the BQ Python API rather than subprocessing out to a shell with bq; it's got much better but it's still a bit Java-esque, and these are examples of times it's a tougher sell than constructing a string a sending it to bash. As ever, lmk if the API isn't designed for these cases and you'd encourage users to use bash.

*this could also be a dict, though less of an imperative.