17  Introduction

Agentic AI refers to LLM-based systems where the pipeline flow is not static or predetermined. Given a query, the system can reason and choose one or multiple courses of action, including using one or multiple tools to achieve its goals.

For example, an agentic AI system can readily answer many factual questions due to the massive datasets used to train it. However, that knowledge is limited to a subset of what was available at the time of training.

Yao et al introduced the ReAct, Reasoning and Acting, framework (Yao et al. 2023), which we will be exploring briefly in Chapter 19.

Yao, Shunyu, Jeffrey Zhao, Dian Yu, Nan Du, Izhak Shafran, Karthik Narasimhan, and Yuan Cao. 2023. “React: Synergizing Reasoning and Acting in Language Models.” In International Conference on Learning Representations (ICLR).