Why AI Agents Need Local Computers | Sarah Sachs

AI agents are transitioning from simple assistants into complex, deterministic processes that require robust infrastructure. By shifting specific tasks from GPU-intensive LLMs to CPU-based execution, organizations can achieve greater reliability, observability, and cost-efficiency. Notion leverages Vercel’s infrastructure to support these deterministic workflows, enabling agents to utilize local file systems and Python scripts for more effective information processing. As agents evolve toward collaborative systems, the primary challenge lies in managing "agent sprawl" through a centralized, intelligent context layer that captures the underlying logic of business operations rather than just raw data. Sarah Sachs, Head of AI at Notion, emphasizes that true enterprise value stems from providing this shared context, allowing companies to remain model-agnostic while ensuring agents can effectively collaborate and execute high-level knowledge work across diverse organizational workflows.