Claude AI

Understanding Claude AI, Claude Code, and the vocabulary that comes with them

Section 1

What Is Claude AI?

Claude is a family of large language models (LLMs) built by Anthropic, a safety-focused AI research company. Unlike a search engine that looks up answers, Claude generates responses by predicting what text is most useful given the conversation so far.

At its core, Claude is trained to be helpful, harmless, and honest — a set of values Anthropic calls Constitutional AI. This means Claude tries to give accurate, thoughtful replies while refusing requests that could cause harm.

Claude comes in several tiers, each balancing speed and capability:

  • Haiku — fastest, lightest; great for quick tasks and high-volume use cases.
  • Sonnet — the everyday workhorse; strong reasoning at practical speed.
  • Opus — the most capable tier; best for complex reasoning, writing, and analysis.

You can use Claude through the web at claude.ai, via the API, or embedded in third-party apps. It handles tasks ranging from drafting emails and summarizing documents to writing code, tutoring students, and analyzing data.

Key Facts
  • Developer: Anthropic (founded 2021)
  • Type: Large Language Model (LLM)
  • Access: claude.ai, API, mobile apps
  • Design goal: Helpful, Harmless, Honest
  • Current family: Claude 4.x (Haiku, Sonnet, Opus)
  • Context window: Up to 200 000 tokens
What Claude Can Do
Write & edit text
Generate code
Summarize documents
Answer questions
Analyze data
Teach & tutor
Brainstorm ideas
Debug errors
Section 2

What Is Claude Code?

How Developers Use It
  1. Open a terminal inside your project folder.
  2. Run claude to start an interactive session.
  3. Ask it to add a feature, fix a bug, or explain a file.
  4. Review the proposed changes before they are applied.
  5. Accept, modify, or reject — you stay in control.
Built-In Capabilities
  • Read and edit files across the repo
  • Run shell commands and tests
  • Search the web for current info
  • Manage git commits and branches
  • Connect to external tools via MCP

Claude Code is Anthropic's official command-line interface (CLI) that brings Claude's intelligence directly into a developer's terminal and code editor. Instead of switching to a chat window, you interact with Claude from inside your project.

Where the regular Claude chat is a conversation, Claude Code is an agentic coding assistant: it can read your entire codebase, write and modify files, run tests, execute shell commands, and propose pull requests — all in one workflow.

Claude Code is designed around a key principle: you remain in the driver's seat. Every significant action (editing a file, running a command) requires your approval unless you explicitly grant broader permissions. This keeps the human in control of the actual repository.

It integrates natively with popular editors:

💻
VS Code
First-class extension
🧠
JetBrains IDEs
IntelliJ, PyCharm, etc.
⌨️
Terminal
Any shell on Mac/Win/Linux
Section 3

Key Terms & Definitions

A reference glossary for Claude AI and Claude Code vocabulary.

Claude AI Terms

A type of AI model trained on massive amounts of text that learns statistical patterns in language. Given an input (a prompt), it generates the most statistically likely useful continuation. Claude is an LLM built by Anthropic.

The text input you send to Claude. A well-crafted prompt gives Claude the context, constraints, and goal it needs to produce a useful response. Prompt engineering is the practice of refining prompts to improve output quality.

The basic unit Claude processes. A token is roughly 3–4 characters or about ¾ of a word. "Hello world" is 2–3 tokens. Models have a maximum context window measured in tokens (Claude supports up to 200 000).

The maximum amount of text (measured in tokens) that Claude can "see" at once — including everything in the conversation so far plus the current message. Information outside the context window is not visible to the model.

Anthropic's training approach that uses a set of written principles (a "constitution") to guide the model's values. Instead of relying entirely on human feedback for every case, the model is trained to critique and revise its own outputs against those principles.

When an LLM generates plausible-sounding but factually incorrect information. This happens because the model predicts likely text rather than looking up verified facts. Always verify critical information from authoritative sources.

Instructions set by the developer (or application) before the user conversation begins. The system prompt tells Claude its role, tone, and constraints — for example, "You are a helpful teaching assistant for a Python class."

A parameter (0–1) that controls how "creative" or "random" Claude's output is. Low temperature (near 0) produces more deterministic, predictable responses. High temperature produces more varied, creative — but potentially less accurate — output.

Claude Code Terms

An AI that takes sequences of actions — reading files, running commands, calling APIs — to accomplish a goal, rather than simply generating text in a single response. Claude Code operates agentically inside your codebase.

A text-based interface where you interact with software by typing commands in a terminal. Claude Code's CLI lets developers chat with Claude directly from their project directory without leaving the terminal.

The ability of an LLM to call external functions or APIs to accomplish tasks. In Claude Code, built-in tools include reading/editing files, running bash commands, searching the web, and managing git. Developers can also define custom tools.

An open standard that lets Claude Code connect to external data sources and services (databases, APIs, internal tools) through a common interface. MCP servers act as adapters that expose capabilities Claude can call as tools.

A special markdown file placed at the root of a project that Claude Code reads automatically at the start of every session. It typically documents the project structure, coding conventions, and workflow instructions so Claude always has relevant context.

Shell commands configured in Claude Code's settings that run automatically in response to events — such as before a file is saved or after a command completes. Hooks let teams enforce linting, testing, or notification workflows consistently.

Special shortcut commands typed with a leading slash (e.g. /help, /clear, /review) that trigger built-in or custom skills inside Claude Code without writing a full natural-language prompt.

A feature of the Anthropic API that stores processed prompt prefixes (like a large CLAUDE.md or codebase context) so they don't need to be re-processed on every request. This significantly reduces cost and latency for repeated, context-heavy interactions.