Hierarchical agents

- Think of hierarchical agents like a company. There are high level agents that organize a problem into smaller goals and oversee lower level agents that execute those tasks and provide progress reports. If it's a complex system, there could be multiple agent levels. The advantage of this approach is that instead of having one agent do everything, it creates resource efficiency by assigning tasks to the best agent and avoids duplicating effort. The hierarchical agent facilitates information sharing, communicates between levels and provides feedback to the rest of the agents. The higher level agents also learn along the way which agents are better at what tasks, and thus continue to simplify the problem and support agent learning across the system. The example of the AI travel agent used at the start of the course is a type of hierarchical agent. It can keep adding new, lower level agents to take on more complex travel requests, rather than be reprogrammed from scratch. This flexibility is one of the biggest advantages of hierarchical agents. New modules can be added or modified to handle changing requirements. It's also resource efficient, as each agent specializes in a specific task, rather than doing multiple ones. Hierarchical agents also take a collaborative approach to decision making. Higher level agents receive feedback from lower level agents. It leads to more informed comprehensive decisions. The collaborative process improves quality and accuracy in decision making. But keep in mind some of the limitations of hierarchical agents. The predefined structure and modules may limit adaptability, which may hinder the agent's ability to adjust. They also follow a top down control flow, which can create bottlenecks if there's a problem with lower level tasks. And hierarchical agents rely heavily on information flow between different levels, so gaps or delays can impact decision making. Lastly, developing hierarchical agents requires significant upfront development and integration. The process involves defining the hierarchy, creating modules and agents for each task, and ensuring seamless communication between levels. To get started with hierarchical agents, look for management situations where it can take on the load of complex decision making processes from a human, such as customer engagement and service, supply chain management or talent management. Look for bottlenecks in your processes that prevent you from scaling operations and investigate how hierarchical agents could take on the work and even do a better job than their human counterparts. Phew, we've looked at six different AI agents, simple reflex, model-based, goal-based, utility-based, learning-based and hierarchical agents. Each has their advantages and limitations, and all create business value in different ways. With this knowledge in mind, let's start experimenting and implementing Agentic AI.