arxivcs.AI2026-07-14
PM-Bench: Evaluating Prospective Memory in LLM Agents
A significant challenge in agentic AI is prospective memory: the ability to execute an intention at a specific future cue or state while other activities are ongoing. We introduce PM-Bench, a text-based benchmark for measuring prospective memory capabilities in modern LLM agents.…