Building AI Agents that actually automate Knowledge Work - Jerry Liu, LlamaIndex

In this talk, Jerry Liu, co-founder and CEO of LlamaIndex, discusses building AI agents to automate knowledge work, particularly focusing on unstructured data like documents. He categorizes agents into assistive (chat-based information retrieval) and automation types (routine task automation), emphasizing the need for a robust document toolbox beyond basic RAG, including data connectors, parsing, and indexing. Liu highlights the challenges of complex documents and introduces LlamaIndex's managed platform with new Excel capabilities for handling un-normalized spreadsheets. He also touches on agent design patterns, contrasting constrained and unconstrained architectures, and presents real-world use cases such as financial due diligence, enterprise search, and technical data sheet ingestion, showcasing how these agents can significantly improve efficiency and decision-making in various industries.