I just don't see how she missed in her example that the post to linear graphql endpoint needs the model to load the graphql definitions, there's no way it's 65x the tokens. Whatever overage it actually is, it's well worth not having to muck around with graphql.