A new 130-page survey explores the concept of "always-on agents," defining them as AI systems whose future actions are influenced by durable state accumulated across interactions. This state is treated as more than simple memory, encompassing elements like task ledgers, permissions, credentials, commitments, provenance, triggers, and externally committed effects. The survey evaluates these state components across six dimensions—authority, scope, mutability, provenance, recoverability, and actionability—throughout their lifecycle, including writing, retrieval, forgetting, auditing, and rollback. AI
IMPACT This survey provides a framework for understanding and building more sophisticated AI agents capable of managing complex, durable states.
RANK_REASON The cluster describes a survey paper on AI agents. [lever_c_demoted from research: ic=1 ai=1.0]
Read on X — Omar Sanseviero (HF research) →
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
