Glossary
A reference of important terms you’ll encounter throughout this course and in the broader AI agent ecosystem.
Agent: An AI system that can autonomously take actions to accomplish a goal. Unlike a chatbot, an agent uses tools, makes decisions, and loops until a task is complete.
API (Application Programming Interface): A way for software programs to communicate with each other. AI platforms provide APIs so developers can integrate AI into their applications.
Auto-Memory: A feature in some AI platforms where the AI automatically saves important information from conversations to remember in future sessions.
Chatbot: A basic AI program that responds to messages in a conversational way. Unlike an agent, a chatbot typically handles one question at a time without using tools, making decisions, or looping through multi-step tasks.
CLAUDE.md: A configuration file used by Claude (Anthropic’s AI) to store persistent instructions, preferences, and rules. Placed in a project’s root directory.
Context Window: The maximum amount of text an AI can “see” at one time. Think of it as the AI’s working memory or desk space. Larger context windows can handle more information.
Configuration File: A file (usually .md or .txt) that contains persistent instructions for an AI agent. Read at the start of every conversation to maintain consistent behavior.
Custom GPT: A ChatGPT feature that lets you create a specialized version of ChatGPT with custom instructions, knowledge, and capabilities. Creating Custom GPTs requires a paid plan (Plus, Team, or Enterprise). Free users can use Custom Instructions or browse existing GPTs from the GPT Store.
Custom Instructions: A free ChatGPT feature that lets you set persistent preferences and instructions that apply to all your conversations. The free-tier alternative to Custom GPTs.
Few-Shot Prompting: Including a few examples in your prompt to show the AI what kind of output you expect. More examples generally produce more consistent results.
Fine-Tuning: Training an AI model on specific data to specialize its behavior. This is a developer-level technique, not covered in this course.
Grounding: Constraining an AI agent to only answer based on information you provide (such as a FAQ document or knowledge base), rather than generating responses from its general training. Reduces hallucinations and makes agents more reliable for customer-facing use.
Gem: Google Gemini’s equivalent of a Custom GPT. A reusable AI agent with custom instructions and personality. Creating Gems requires Gemini Advanced (paid plan).
Hallucination: When an AI generates information that sounds plausible but is factually incorrect. Always verify important facts from AI output.
LLM (Large Language Model): The AI model that powers agents. Examples include Claude, GPT (behind ChatGPT), Gemini, Llama, and Mistral. The “brain” of an AI agent.
Local Model: An AI model that runs on your own computer rather than in the cloud. Ollama and LM Studio let you run local models. Benefits: privacy, no per-use cost. Drawback: requires decent hardware.
Multimodal: The ability of an AI to understand and process multiple types of input, including text, images, audio, and video. Gemini and ChatGPT both support multimodal input.
MCP (Model Context Protocol): A standard for connecting AI agents to external tools and data sources. Allows agents to interact with files, databases, APIs, and more.
Memory (Agent): Information that persists between conversations. Includes configuration files, auto-memory, conversation history, and skills.
Modelfile: Ollama’s configuration file format. Defines which model to use, system prompt, and model parameters.
Observe-Think-Act Loop: The core cycle every AI agent follows: observe (gather information), think (decide what to do next), and act (execute an action). The loop repeats until the task is complete. Also called the “core agent loop.”
Prompt: The input text you provide to an AI. Can be a question, instruction, or any text the AI should respond to.
Prompt Engineering: The practice of crafting effective prompts to get better AI output. A core skill for building agents.
Prompt Injection: A security concern where malicious instructions are hidden in input to trick an AI into doing something unintended. A reason to be cautious with agents that process external input.
RAG (Retrieval-Augmented Generation): A technique where the AI retrieves relevant information from a knowledge base before generating a response. Not covered in this course but important for advanced agent systems.
Reverse Prompting: A technique where you ask the AI to ask you clarifying questions before it begins a task. Improves output quality.
Self-Modifying Prompt: A pattern where the AI updates its own configuration file with new rules learned from mistakes. Over time, this reduces errors and improves consistency.
Skill: A standardized, reusable set of instructions that makes AI output more consistent and deterministic. Can be shared between users and platforms.
System Prompt: Hidden instructions that define an AI agent’s behavior, personality, and constraints. Set before the conversation begins and applies to all interactions.
Temperature: A setting that controls how creative/random an AI’s output is. Lower temperature = more predictable. Higher temperature = more creative. Default is usually fine for most tasks.
Token: The basic unit of text that AI models process. Roughly 1 token equals about 4 characters or 0.75 words in English. Pricing and context limits are measured in tokens.
Tool: A capability that an agent can use to interact with the outside world: reading files, searching the web, running code, calling APIs, etc.
Zero-Shot Prompting: Giving the AI a task without any examples. Works well for simple tasks but may need few-shot prompting for complex or specific output formats.