Understanding AI Agents: Tools, LLMs, and Memory
- Agentic AI
- Enterprise h2oGPTe
- H2O Driverless AI
This lesson breaks down what separates AI agents from generative AI and RAG, and explains the three core components that make an agent work: tools, a large language model, and memory. It walks through concrete examples of each, from Python and web browsing tools to vector database memory, and introduces the spectrum of agency, from a single no agency recommendation to an agent that resolves a task on its own. The lesson closes by placing Enterprise h2oGPTe at the full agency end of that spectrum, able to complete tasks autonomously.
Key takeaways
- The three components of an AI agent are tools, a large language model, and memory
- How AI agents differ from generative AI, which produces content, and RAG, which retrieves context for generation
- What the spectrum of agency looks like, ranging from a single no agency recommendation to full autonomous task completion
- The tool examples given for h2oGPTe, including Python and Bash execution, web browsing, file handling, and Driverless AI modeling
- Why memory lets an agent learn from past interactions, often implemented as a vector database storing customer history
- How h2oGPTe sits at the full agency end of the spectrum, completing tasks autonomously without human intervention
Start owning your AI