# A Better Way to Work with AI Agents By:: [[Brian Heath]] 2024-04-23 To utilize AI agents better, one must consider them incapable of performing many tasks in one fluid motion. In other words, one needs to not think of them as human. When one speaks with a human customer service representative, one engages with a person who performs many functional tasks simultaneously. They look up data, respond with appropriate language, follow a script, know when to extend beyond the script, and constantly correlate all of this dynamic information into a coherent narrative and set of action items. This is so typical that one rarely pays attention to the complexity of essential human communication. Unfortunately, our tendency to humanize entities means we also think AI agents or chatbots should respond the same way. But this is different from how they generally work. They are primarily programmed to excel at doing a few things, and rarely, if ever, does this combine to create an experience one would have with a human. So, stop thinking of AI agents as being able to replicate the capabilities of humans. Instead, think of how one might chain a series of AI agents to perform the tasks typically done automatically by a single human brain. To do this, one must contemplate and break down the functions required to do the work at a fundamental level beyond what is typically done in "scientific management" practices. For example, a customer service job manual might recommend a way to answer the phone and the steps required to do a data lookup request. Still, it rarely identifies ways to delight the customer's day that humans can do in unexpected ways. Suppose one sets up an AI chatbot as a customer service representative and expects it to do everything from a long list of instructions. In that case, one will spend a long time trying to get the bot to behave like a below-average customer service representative. However, one should not be restricted to only using one AI agent. Programming a series of AI agents with specific tasks that collaborate is much easier. This closer mimics the flexibility of the human brain. One agent keeps track of the caller's tone and ensures appropriate responses. Another agent performs the data look request while another monitors all of the information and looks for deeper topical lines of questioning to discern if there are other underlying issues. Finally, another agent might be the conductor to ensure it all comes together appropriately. This approach is easier to specify and reflects a more accurate paradigm of AI's current existence. #### Related Items [[Artificial Intelligence]] [[Methodology]] [[Work]] [[Process]] [[Function]] [[Paradigms]]