In a commentary for Insight Jam, Pocket Prep’s Michelle Marlowe argues that engagement is not learning and that AI is widening the gap, noting that engagement metrics like streaks and completion rates are "the industry’s empty vanity metric." Researchers have named the gap AI opens the "learning-performance paradox," which Hassan Khosrau and a team of learning scientists describe as "the well-documented phenomenon whereby AI tools can enhance short-term task outputs while simultaneously undermining durable learning." In a 2025 study, Hamsa Bastani and colleagues gave nearly 1,000 high school math students access to AI tutors built on GPT-4 in two versions. One version, "GPT Base," simply handed over answers; while students had access, their performance soared 48%, but when the researchers took it away, those students scored 17% worse than classmates who never used it. The second version, "GPT Tutor," prompted to give teacher-designed hints instead of full-blown answers, drove even larger in-session gains (127%), yet largely erased the penalty once the tool was gone. Khosrau points to "metacognitive laziness, in which learners abdicate the cognitive and metacognitive effort required for deep understanding," essentially a learning penalty from cognitive offloading.