The rise of AI in education has led to significant advancements, promoting automated grading to reduce educator workload and to enhance pedagogy. However, its integration raises complex pedagogical, ethical, and technical questions. This systematic review examines the intersection of automated grading tool development and educational assessment through the lens of the activity theory. Our analysis, informed by literature since 2010, reveals a critical need for comprehensive evaluation frameworks addressing the iterative nature of technology development and interdisciplinary collaboration. Key breakpoints in existing studies include oversight of the reliability and validity of assessments, ethical considerations, coherent evaluation rules, interdisciplinary collaboration, and agentive and constructive roles of users. Addressing these issues requires a holistic approach that bridges technical and educational perspectives, fostering trust and supporting meaningful learning outcomes. Enhanced collaboration and ongoing professional development are crucial for creating AI-driven assessments.