The article explains how to identify agentic AI use cases, focusing on the concept of "agents" powered by generative AI, particularly large language models (LLMs). Agents automate tasks that traditionally resist automation, such as handling support tickets or creating marketing campaigns. For an application to qualify as an agent, it needs a clear task, the ability to use specific tools, and some level of reasoning to determine actions based on inputs. Three main concepts define an agent: tasks, tools, and reasoning. Tasks have a distinct beginning and end, triggered by inputs like chat messages or alerts. Tools distinguish agents from standard LLMs; they allow agents to interact with external information sources, such as databases or the internet. The article categorizes tools into three types: context fetchers (for data retrieval), persisting actions (for executing commands), and orchestration (for managing multiple agents or workflows). Reasoning enables agents to dynamically select the appropriate tools and actions needed to complete their assigned tasks. To identify potential agent use cases, the article suggests compiling a list of processes within teams or organizations that could benefit from automation. It provides guiding questions to help users identify tasks that are repetitive, time-consuming, or bottlenecked by subject matter experts. The proposed Agent Identification Framework includes five key characteristics: input/trigger, objective, scope, tools, and guidelines for the agent. The framework consists of two steps: documenting examples of agent runs and extracting the five characteristics to clarify the agent's role and responsibilities.