Researchers have developed CogniFold, a brain-inspired memory system for autonomous agents. Unlike traditional reactive and retrieval-based memories, CogniFold continuously organizes experience into cognitive structures through self-emerging graph-topology organization. This allows the agent to autonomously assemble, merge, decay, and relink knowledge as it accumulates. The authors claim that CogniFold achieves this through a prefrontal intent layer, which enables intentional control and decision-making. They evaluate their system using various benchmarks, demonstrating its ability to produce memory structures matching cognitive expectations and concept emergence. --- Why it matters: This matters because CogniFold could enable the development of more autonomous and proactive AI assistants that can learn from experience and adapt to new situations without relying on human intervention. Source: https://arxiv.org/abs/2605.13438