Agentic Data Environments

Abstract

Autonomous agents are shifting data systems from read-only analytics toward consequential, state-mutating workflows. This position paper introduces Agentic Data Environments (ADEs) - active data-system substrates that prepare task-relevant information while governing how agents explore, modify, and propagate state. We outline three information management directions — Agentic Information Management, Agentic Information Retrieval, and Agentic Data Elicitation, and two safety primitives - state branching and Data Flow Control. Together, they aim to amplify agent capabilities while bounding the risks of autonomous action across databases, filesystems, processes, APIs, and external services.