What Are AI Agents?
Human agents, say a travel agent or real estate agent, perform tasks on behalf of their clients and are useful because of the specialized training they can bring to bear. For example, a customer may enlist the help of a travel agent to take a month-long trip to Africa. The customer tells the agent about what she wants to see and do and provides parameters, such as dates for travel and a budget. Then the agent works to set up the trip, reports back with options, and ultimately books the trip for the customer. AI agents conceptually work the same way. A human asks for a task to be completed. She provides details and parameters under which the agent will operate, and then the agent autonomously does the work and reports its results.
AI agents will typically use a powerful LLM to understand what the requester wants, and as it does, it will devise a plan to complete the task. Using RAG, the AI agent will get the information it needs, and it may request other information as needed to complete each part of its plan. The agent retains data on how it completed previous tasks, effectively giving it memory and the possibility to improve its process. An AI agent can also use tools such as ML models, automated robotic processes, and other LLMs—sometimes less powerful but more focused—to complete its task. It’s up to users to set parameters, such as when to pause for approvals before proceeding.
The benefit of AI agents is their ability to complete complex tasks autonomously and to leverage data, including that which describes company polices and procedures, as they go. As such, agents offer the potential of substantial improvements in productivity, speed, and accuracy.
AI agent use cases are myriad and include software development, customer support, order taking, inventory replenishment, HR benefits support, spontaneous cyberthreat response, and executive decision support.
What Are the Key Differences Between AI, GenAI, ML, and AI agents?
| What Is It? | What Does It Do? | What Does It Require? | |
|---|---|---|---|
| Artificial Intelligence (AI) | It’s the overall term for a broad discipline of computer science going back decades. | It solves problems that are traditionally addressable by humans but difficult for computers. | AI can run on hardware as varied as embedded systems, phones, computers, and cloud clusters. |
| Generative AI (GenAI) | It uses large neural networks and extensive training to create models that generate new content. Text generators are known as large language models, or LLMs. | It can write text and create images and sounds that appear to be created by humans. | Large clusters of GPU-equipped servers take extensive time and data to create the models. |
| Machine Learning (ML) | It uses smaller neural networks and curated, categorized data to learn to perform a single function. | It can spot patterns and make predictions when provided simple or moderately complex data. | Training is sped up by GPUs. AI inferencing, can be done on most CPUs; there’s no need for specialized chips. |
| AI Agents | They use powerful LLMs to understand and complete complex tasks for humans. Agents use tools and external data and can improve as they complete more tasks. | They autonomously provide services for humans. Tasks performed include first-level customer support and HR benefits support. | They use existing LLMs as a foundation. Agents work by building environments where LLMs have what they need to provide their services. |
link