Skip to content

Glossary

Key terms used across this blog, focused on AI development and generative AI.


AI Fundamentals

AI (Artificial Intelligence)

The broad field of building software that mimics human cognitive abilities such as reasoning, learning, and problem-solving.

Generative AI

AI that creates new content — text, images, audio, or code. ChatGPT and Claude are leading examples. Unlike traditional AI (classification, prediction), its primary purpose is to produce.

LLM (Large Language Model)

A neural network trained on massive text corpora to understand and generate natural language. GPT-4, Claude, Gemini, and Llama are prominent LLMs.

Token

The smallest unit an LLM processes. In English, one word is roughly one token. API pricing and context window sizes are measured in tokens.

Hallucination

When an AI generates plausible-sounding but factually incorrect information. A key reason why fact-checking AI outputs is essential.

Fine-tuning

Additional training of an existing LLM on domain-specific data to specialize it for particular tasks.


Prompt

The text input given to an AI — instructions, questions, or context. Writing effective prompts is the foundation of AI usage.

Prompt Engineering

The practice of designing and optimizing prompts to get desired outputs from AI. Techniques include role-setting, few-shot examples, and constraint specification.

System Prompt

A special prompt that defines the AI's behavior, rules, and persona, separate from the user's input.

Context

All information the AI considers when generating a response — conversation history, system prompt, attached files, and more.

Context Window

The maximum number of tokens an LLM can process at once. Claude Opus 4 supports approximately 1 million tokens. Larger windows enable handling longer documents and more complex tasks.


Agents & Automation

AI Agent

An AI program that autonomously plans, uses tools, and evaluates results to accomplish goals. Unlike simple Q&A, agents execute multi-step workflows independently.

Sub-agent

A child agent that a main agent delegates specific subtasks to. Used for parallel processing and context isolation. In Claude Code, launched via the Agent tool.

Multi-agent

An architecture where multiple agents collaborate on tasks. Each agent has different expertise, enabling efficient division of labor for complex work.

Tool Use / Function Calling

The ability of an LLM to invoke external tools (search, computation, file operations, API calls) to retrieve or manipulate information. The core mechanism that enables agents.


Claude Code

Claude Code

Anthropic's CLI-based AI development tool. Enables interactive AI-assisted coding directly from the terminal or editor.

CLAUDE.md

A configuration file that gives Claude Code project context — coding conventions, directory structure, rules, and constraints.

Hooks

Shell commands that execute automatically in response to Claude Code events (before/after tool execution, on notifications, etc.). Used for guardrails, auto-formatting, and workflow automation.

Skills

Definition files that extend Claude Code's capabilities. Triggered by specific keywords, they provide specialized knowledge and workflows. Stored in .claude/skills/.

Harness Engineering

The systematic design and optimization of AI tool configurations — skills, hooks, prompts, and settings (in Claude Code, Copilot, etc.). Like fitting a harness to a horse, it maximizes the AI's effective output. A superset of prompt engineering.

Context Engineering

The intentional design of the context fed to an AI (CLAUDE.md, memory, file references, etc.) to improve output quality. A component of harness engineering.


Protocols & Integration

MCP (Model Context Protocol)

An open protocol by Anthropic that standardizes how AI applications connect to external data sources and tools. Often described as "USB-C for AI."

API (Application Programming Interface)

An interface for software-to-software communication. In AI, typically refers to using LLMs programmatically via the Claude API, OpenAI API, etc.

RAG (Retrieval-Augmented Generation)

A technique that retrieves relevant information from external databases and feeds it as context to an LLM, enabling answers grounded in up-to-date or domain-specific knowledge.


Development Tools

GitHub Copilot

GitHub/Microsoft's AI coding assistant. Integrates into IDEs for code completion and chat-based support.

Codex CLI

OpenAI's CLI-based AI development tool. A terminal-integrated tool in the same category as Claude Code.

Gemini

Google DeepMind's LLM. Strong integration with Google services.


Quality & Safety

Guardrail

Safety measures that prevent AI from producing inappropriate outputs or executing dangerous operations. Includes automated checks via Hooks and constraints in system prompts.

Alignment

Ensuring AI behavior matches human intent and values — training and designing AI to be helpful, honest, and harmless.


This glossary is updated regularly. Feedback on missing terms is welcome.