AI evaluations, crucial for determining if an AI feature is shippable, often overlook the critical role of designers despite fundamentally addressing user experience questions. These evaluations, typically driven by engineers and product managers, risk shipping products that are technically sound but fail users because designers are not involved early enough in defining criteria. Designers bring invaluable taste and judgment that quantify complex user needs, discerning what constitutes a "right answer" beyond mere metrics. They are adept at handling scenarios where AI lacks information, applying their experience with error messages and dead ends to guide users effectively. Furthermore, designers understand the complexities of real user behavior, making them essential for creating accurate test scenarios that synthetic users cannot replicate. They also challenge underlying assumptions in evaluations, ensuring that business goals align with genuine user needs rather than just ease of measurement. Lastly, many AI evaluation failures are, at their core, design problems that require user-centered solutions rather than just prompt adjustments. Integrating designers into the AI evaluation process from the outset ensures user-centric criteria and ultimately leads to more successful AI products.