Peer review assessments in large-scale online courses face challenges around inconsistency, delays, varying feedback quality, and concerns over the expertise of student graders. While peer assessment can be implemented at scale, providing efficient and effective feedback remains difficult. To address these issues, the team at Coursera, a leading online education platform, implemented an AI Grading system leveraging generative artificial intelligence (GenAI) to provide immediate, consistent, and scalable feedback aligned with instructor rubrics. An initial beta test graded approximately 300,000 text submissions across several courses. Key metrics showed the AI system provided feedback 45 times faster than human grading, with 90% of learners satisfied with the AI feedback. Interestingly, first attempt pass rates were lower (72% versus 88%) and average grades were 3% lower than human grading, suggesting increased rigor. At the same time, course completions rose 16.7% with faster AI grading, a promising sign of increased learner engagement and persistence. While showing promise for scalable assessment, open questions remain around balancing rigor and accessibility, supporting learner adaptation, ensuring feedback relevance across subjects, and determining which assignments still benefit from human collaboration. The results demonstrate AI’s potential to enhance assessment scalability and speed while providing rich data to inform improvements. Careful balancing of AI and human expertise will continue to be paramount as further AI integrations are developed for online courses.