Computer-based grading tools have existed for nearly as long as computing courses. The majority of these tools have focused on completely automatic grading of functional requirements, leaving no room for subjectivity, and generally eschewing human feedback in favor of total automation. We argue that these tools are of little practical use because they severely limit the types of assessments that can be graded, and force the user to adopt to the paradigms of the grading tool, rather than vice versa. We present Agar, a tool designed to compensate for those possible shortcomings, discuss the design of Agar, and discuss unanticipated usage patterns that have already sprung up in our user base.