Researchers published a paper on the net value of agent memory. They found many memory systems underperform simple retrieval from raw history. They introduced the terms memory benefit and memory tax to describe gains and losses. Memory tax arises when memory changes remove or obscure evidence. The paper studies how memory tax emerges from system and evaluation design. The authors generalize the concept to broader memory use cases. They will open-source the code and results. The work aims to shift focus from raw performance to net memory value. It offers a new lens for designing and evaluating memory systems.