Intelligence Is Free, Now What? Data Systems For, Of, and by Agents
With AI intelligence becoming abundant and inexpensive, the bottleneck shifts from models to data systems. Autonomous agents need new data infrastructures designed for agents, made of agent-generated data, and managed by agents themselves. The post examines key challenges in building these next-generation systems for scalable agent deployment.
Background
The BAIR (Berkeley Artificial Intelligence Research) blog post discusses a turning point where AI reasoning ("intelligence") has become dramatically cheaper, shifting the bottleneck from model capability to the data infrastructure needed to manage and trust AI agents. The post was published in July 2026, reflecting on advances since GPT-4 and similar large language models. Key context: AI agents (autonomous programs that plan and execute multi-step tasks) are proliferating, creating new problems around data provenance, security, and coordination. The authors are likely UC Berkeley researchers. The "Now What?" framing echoes a broader industry debate about whether raw model intelligence has been "solved" enough that practical deployment challenges—data pipelines, tool-use, verification, agent-to-agent communication—are now the frontier. The piece distinguishes three roles for data systems in an agentic world: systems built *for* agents (infrastructure), systems built *of* agents (data generated by agents), and systems built *by* agents (agents that manage data themselves).