What is Long-Horizon Agent Memory?
Why AI Agents Forget, and Why the Fix Isn’t More Storage
TL;DR
AI agents forget during long tasks not because the context window is too small, but because standard memory retrieval finds semantically similar information rather than causally relevant information.
AMA-Bench, the first benchmark built from real agentic trajectories, confirmed this causality gap. The AMA-Agent system addresses it with a causality g…
