Intelligence Is Free, Now What? Data Systems For, Of, and by Agents (bair.berkeley.edu)

The rapid decline in the cost of AI capabilities, now approaching nearly free inference, marks a significant turning point for the AI/ML community. As the price for advanced models like GPT-4 drops from $30 per million tokens to under $1—and even below $0.10—this democratization of intelligence facilitates greater accessibility for knowledge work. The authors highlight this transition as an opportunity to rethink data systems that will increasingly support autonomous agents, enhancing how tasks are performed and managed. Three key challenges emerge from this new landscape: the need for data systems designed specifically for agents, to facilitate the unique interactions between agents and data; systems that can manage and coordinate multiple agents to handle complex tasks; and systems that agents can autonomously create and verify. Enhanced mechanisms for querying, memory structuring, and agent communication are crucial. By integrating agentic speculation and tailored query interfaces, data systems can be reimagined to effectively support vast numbers of agents, driving efficiency and innovation in knowledge work, while tackling the inherent complexities of concurrent operations, memory management, and inter-agent communication.