In this monologue podcast, Serhii Sokolenko discusses preparing agents for the "Iceberg Age" by complementing LLMs with factual data from analytical and operational databases. He begins with an example of a chatbot failure, then explains how the industry is evolving to use agents that reason, call tools, and access external APIs and databases. Sokolenko highlights the limitations of LLMs, such as dated knowledge and lossy compression, compared to the vast and up-to-date knowledge in analytical and operational databases. He introduces Apache Iceberg as a standard for storing relational tables and emphasizes the importance of throughput and data access security when using agents with analytical databases. Sokolenko then presents Tower, a serverless Python platform for building agentic workflows, and demonstrates how it can be used to maintain a database of stock information using an agent that intelligently pulls data from external sources. He concludes by discussing the future of Agentech, addressing adoption blockers, and introducing Tower's approach to creating a portable Python runtime.
Sign in to continue reading, translating and more.
Continue
