Connecting an LLM to the agentic system is just the first step. You need to understand how the LLM behaves in practice before you can effectively design the agent around the LLM. That way, there will be no surprises—you'll know exactly what the model is capable of doing, and where it may fall short. You'll then be able to put this assessment to good use by building an agent that takes advantage of the LLM and doesn't waste time or tokens on unrealistic behaviors.

Previously, you focused on connecting the agent to a large language model (LLM) and ensuring it uses that LLM effectively. Now, it's time to build the agent's workflow. This workflow will form an overall execution loop within which the agent will repeatedly reason and act. You need to make sure this loop is constructed properly so that it supports the agent's goals as well as your business objectives for the agentic initiative.

You've built your agent to reason effectively, and not only that, but to maintain an awareness of important context while it reasons. But this is really only half of the agentic equation. The other half is to ensure the agent can take actions in an environment. That's key to turning it into an actual automated system. And, the way you facilitate actions is by providing the agent with tools. So, that's what you'll do in this lesson.

Retrieval-augmented generation (RAG) is a supplemental, yet powerful way of making AI agents even more capable. Many agentic systems incorporate RAG to help mitigate the issue of limited memory and context windows in LLMs. An agent can still review and evaluate key pieces of information without having to be fed all of that information directly. In this lesson, you'll set up your agent for RAG so it can make more informed decisions based on extensive organizational documentation.

An important part of building a capable agent is ensuring that it can perform its assigned tasks within acceptable boundaries. Until you incorporate these boundaries, the agent cannot be relied upon to produce safe and consistent results in a production environment. That's why, in this lesson, you'll employ various techniques to prevent mistakes from having a significant negative impact on the agentic system as a whole.

Agentic AI is, fundamentally, software—and like any software, it must be tested prior to launch. If you're familiar with the world of software development, then you'll probably have some idea of how to approach testing an agent. But, there are also some key elements that distinguish testing an agent from testing a normal application. In this lesson, you'll employ various methods for testing agentic AI to ensure it is meeting expectations.

You've thoroughly developed and tested your agentic system, so naturally, the remaining step is to deploy it. But, deployment is not just a matter of flipping a switch and then walking away—it requires planning and design work, like any other aspect of the development process. You need to choose an interface through which to expose the agent to its users, and you need to make sure you're actually ready to deliver the agent to production instead of creating another prototype. In this lesson, you'll make sure you're truly prepared to deploy your agentic system.

You'll wrap things up and then validate what you've learned in this course by taking the credential exam.