Agents are getting popular, but where do you start? Gartner and others have called 2025 the Year of Agentic AI. I think the key to building agents is similar to building anything else with machine learning (ML) or other forms of AI. First, you need to be an innovator. An innovator sees the future and wants to build it when the tools become ready to do so.
Leonardo da Vinci was an innovator. He was a young apprentice who happened to get a job at a place that had a lot of tools to build the things that his mind could imagine. His boss was also a painter who taught him how to paint. If you're an innovator, then you're the one who needs to see the future that AI agents and machine learning can do for your company. Oracle provides the tools for you to succeed.
Define the Business Need
You need to start with the business needs! Not just the current business needs, but the future business needs as well. You need to consider where the business must innovate to survive and excel. In this blog, I view Oracle's agents as pieces of code that perform helpful actions, such as: I would like an autonomous robot (using code/agent) to do a task that happens to be inside a physical robot of some kind. If I need a robot to vacuum the floor, I need to create an agent/code inside the robot that understands the task with all obstacles it will encounter and how to get through them (and an ORA-600 error, or kinder message, if it encounters an unspecified obstacle). OK, Roomba beat me to that one. But solving the next problem—specific to your company, to automate things such as identifying fraud, finding more customers, HR tasks, various Advisors with information from manuals, and all of those other items that need an autonomous robot to assist someone in your business are soon to be near the top of your to-do list. Oracle Security, the Oracle Cloud, and the Oracle Vector Database in 23ai are here to help you do this fast!
Vector Databases, RAG, and the Power of OCI
If you saw my last blog series on what happened at CloudWorld (it's on the Viscosity website), you saw how we can take words or images and build a numerical representation of their semantic meaning in multi-dimensional space (stored in the Oracle Vector Database). We could then find similar things that matched the image (close by) or didn't match (far away) in that multi-dimensional space. We use Retrieval-Augmented Generation (RAG) to keep our database protected from LLMs (Large Language Models) that want to learn from our data and use an LLM inside the database to get only what we need. We don't want to share that information.
Here, I'm going to focus on an autonomous robot (chatbot) that can answer my own questions using a variety of documents that I put into the Oracle Cloud and leverage the power of Oracle Agents & an LLM. It would be nice if I could chat with my 19c or 23ai documentation (PDFs) inside my own database, as I have questions from time to time. I could also use Oracle's SelectAI or APEX to talk to my database. It's nice to use an LLM like Llama or Cohere to get answers, while also protecting my private documents and database information in the Oracle database from prying eyes. Larry Ellison gave a virtual announcement talking about the Oracle Vector Database when it came out on the 23.4.0.24.05 release of the 
Oracle 23ai Database, and during that virtual talk, he said the Oracle Vector Database would be absolutely free in the Oracle database (the converged database just got better)!
Your First Steps: LiveLabs and GenAI Agents
A great way to learn about agents or the vector database, in addition to getting started in this blog, is to complete one of the Oracle LiveLabs that Oracle has available. It's like going to a virtual class on your own schedule to learn something new. The best part of Oracle LiveLabs is that it's free, it's online, and you can use your own database or cloud to complete different LiveLabs or use Oracle's for free. Did I say that it was free? Don't tell Larry! Just go to https://apexapps.oracle.com/pls/apex/r/dbpm/livelabs/home.
I'll reference a couple of LiveLabs in this blog, but the main one that I recommend is: "Unlock your Enterprise Knowledge: Building your first Oracle GenAI RAG Agent." If you're in any OCI region (Oracle Cloud Infrastructure - or just Oracle Cloud), you can click on Analytics & AI. Then hover over AI Services, and you can use Generative AI and other AI services like Language, Speech, Vision, and others. Every Oracle Cloud data center is identical except for the size. They have the same infrastructure, CPUs, GPUs, RDMA Network, autonomous operating system, and database applications like Fusion Apps, ERP, HCM, Oracle CX (Customer Experience), Netsuite, Oracle Cerner Health, & other industry applications.
A couple of the Oracle Pre-trained Foundation Models in GenAI are listed in the image in this blog are (from the Oracle Docs). You can see all of the LLMs available on OCI in the Oracle docs (subject to change). Using OCI to do things with AI has 
one additional advantage: Security! Oracle's Gen2 Cloud security for both public and private clouds has improved with data security, app security, user identity security (soon adding biometrics), and network security. With the Autonomous Database running on Autonomous Linux, much of the human error is removed. Oracle does the configuration, encryption, backup, recovery, and updates patches with no human labor, which means no human error and time savings. In 2025, Fusion Apps, Cx, Sales & Marketing, HCM, Netsuite, and many industry applications will all be moved to autonomous database using APEX in the Oracle cloud. It's the current direction of Oracle. AI, though, is the biggest focus with Agents leveraging algorithms and LLMs. Oracle currently trains LLMs for Nvidia, Cohere, Llama, and xAI.
If you are in a region that supports "Generative AI Agents" currently (like Chicago), you can click on "Generative AI", but you can ALSO click on "Generative AI Agents" (at the bottom of the AI Service list). We'll use this to create agents using the Llama LLM (Large Language Model) residing on OCI. But first, let's take a look at all of the Generative AI offerings that are available in ALL regions (see image).
The Generative AI offerings include being able to Chat with an LLM, including setting filtering for explicit content. You can use GenAI to generate new information or Summarize an email or document. You can also build Embeddings. These LLMs are being updated with new models (as others are retired) on a regular basis. But this blog will focus specifically on GenAI Agents.
Now let's go from the Generative AI choice (within AI Services) to the Generative AI Agents choice in Part II. Our goal will be to build an Agent that accesses documents that we have and answers questions that we ask the LLM. It does this by taking our question and using the semantic meaning and turning it into vector embeddings, which will then look through our documents (which also are turned into vector embeddings), and then find semantic matches to return a response that is turned back into English. We'll also see that we can use GenAI to take things and turn the answer into another language (Spanish, French, German, etc). We can even have it talk like a pirate if we want.
Now that everything’s set up, the fun begins—stay tuned for Part II, where we’ll take the next step and build your first generative AI agent.
Ready to Take Your Oracle Skills to the Next Level?
Go deeper with Rich’s insights by gaining access to past webinar recordings, plus exclusive entry to upcoming members-only sessions through OraPub’s training community.
Check out Viscosity’s event page for upcoming virtual and on-site training opportunities.
SUBMIT YOUR COMMENT