Aug 25, 2026 Why We Need Long-Term Memory Layers for Agentic Scaling Aug 25, 2026 Non-markovian worlds => long context pre-training and RL. OPD x