Taking you to the card → The case for open weights: strong at agents, weak at science