Databases: The Final Boss of Agents

Large language models (LLMs) are evolving into autonomous agents that can reason, plan, and act over complex systems. We contend that databases pose the hardest and most important challenge for agents, due to their unforgiving correctness and performance requirements. This talk presents research on using agents for databases in two ways. The first part focuses on tuning agents that automatically optimize database configurations, including physical design (e.g., indexes), system knobs, and query execution strategies. The second part shifts to coding agents as collaborators in building database systems themselves. As part of this talk, we will discuss our successes and failures in using agents across both efforts and what is needed to automate them further.