Supporting Our AI Overlords: Redesigning Data Systems to be Agent-First
🤖 AI Summary
Existing data systems struggle to efficiently support high-throughput, heterogeneous, redundant, and controllable speculative data operations—termed “agent speculation”—required by LLM-based agents, leading to query inefficiency and resource waste. This paper proposes an “agent-first” data system architecture that systematically characterizes the four defining properties of agent speculation: scale, heterogeneity, redundancy, and controllability. Leveraging this characterization, we design a novel declarative query interface, an adaptive query processing engine, and a dedicated memory store. By unifying agent behavioral modeling with systems-level optimization, we establish a new paradigm for query execution, dynamic resource scheduling, and state management tailored to agent workloads. Experimental evaluation demonstrates significant improvements in speculation throughput and resource utilization. To our knowledge, this is the first architecture-level solution for agent-native data systems.
Application Category
📝 Abstract
Large Language Model (LLM) agents, acting on their users' behalf to manipulate and analyze data, are likely to become the dominant workload for data systems in the future. When working with data, agents employ a high-throughput process of exploration and solution formulation for the given task, one we call agentic speculation. The sheer volume and inefficiencies of agentic speculation can pose challenges for present-day data systems. We argue that data systems need to adapt to more natively support agentic workloads. We take advantage of the characteristics of agentic speculation that we identify, i.e., scale, heterogeneity, redundancy, and steerability - to outline a number of new research opportunities for a new agent-first data systems architecture, ranging from new query interfaces, to new query processing techniques, to new agentic memory stores.
Problem
Research questions and friction points this paper is trying to address.
Redesigning data systems for AI agent workloads
Addressing inefficiencies in agentic speculation processes
Developing new architectures for agent-first data systems
Innovation
Methods, ideas, or system contributions that make the work stand out.
Agent-first data systems architecture redesign
Leveraging agentic speculation characteristics scale heterogeneity
New query interfaces processing techniques memory stores
🔎 Similar Papers
No similar papers found.
