In this talk I combine philosophy of education and philosophy of technology to demonstrate that the promises surrounding educational AI are neither as plausible or as desirable as they are portrayed in edtech companies’ marketing and in the media.  Proponents of AI-based educational tools claim that they will improve learning gains by tailoring material and delivery to individual students, while also facilitating the work of the teachers by automating routine and unpleasant tasks (like grading, lesson planning or report writing). However, I demonstrate that many commercial tools available today are built on questionable assumptions about pedagogy and good education. The most dominant student-facing tools today – Intelligent Tutoring Systems – frame learning primarily as acquisition of knowledge and ignore the social and experiential aspects of learning. They are thus unlikely to bring the overall positive impacts promised by edtech developers. In turn, I argue that teacher-facing technologies are likely to replace one kind of busywork with another. They suffer from a narrow framing of which aspects of teaching are worthwhile, vastly overestimate AI’s ability to perform teachers’ tasks independently, and involve additional burdens that are rarely considered by the proponents of the technology.