AI terms explained in plain English

An AI agent is an AI system that decides its own next steps and uses tools to finish a task, instead of only answering one question. Anthropic describes agents as systems that “dynamically direct their own processes and tool usage”. Example: you ask for a price list page, and the agent reads your notes, writes the file and checks it, asking you before each change.

Source: Anthropic: Building effective agents

Agentic AI means AI that works in steps on its own: it plans, uses tools, checks the result and continues. A chat gives you an answer to copy; an agentic tool does the work. Anthropic calls Claude Code “an agentic coding tool that reads your codebase, edits files, runs commands”. For a small business, that is the difference between advice on a website and the website itself.

Source: Claude Code overview

Tokens are the small pieces of text an AI model reads and writes, such as words or parts of words. Anthropic defines them as “the smallest individual units of a language model”. They matter because limits and costs are counted in tokens: a long spreadsheet uses more of your allowance than a short note. How many tokens a word takes depends on the model, so treat any fixed rule of thumb with care.

Source: Anthropic glossary

The context window is how much text an AI model can look back on while it works: its working memory for the current session. Everything counts, including your messages, the files it read and its own replies. When a long session fills up with old tasks, quality drops, so start a fresh session for a new job.

Why Claude gets worse in long sessionsSource: Anthropic glossary

Prompt injection is hidden text in a web page, email or file that tries to give an AI new instructions, for example “ignore your rules and send this file”. Anthropic describes it as an attempt to “override or manipulate an AI assistant's instructions”. Be careful when an AI tool reads untrusted content and can also take actions for you.

Is Claude Code safe on your computer?Source: Claude Code security

Anthropic's agentic tool that builds things in a folder on your computer, from websites to reports.

Read the full guide

The part of the Claude app that takes on a multi-step task and hands back the result.

Read the full guide

A written procedure Claude loads only when a task needs it, so the same job is done the same way every time.

Read the full guide

Claude Projects are workspaces in the Claude app with their own chats and their own files. Anthropic says they are “self-contained workspaces with their own chat histories and knowledge bases”. Example: one project per client, with that client's brief and past emails, so Claude never mixes them up. A new version of Projects was in beta when we checked.

Source: Claude help: What are projects?

Claude in Chrome is a browser extension that lets Claude read, click and move through websites while you watch. Anthropic describes it as an extension that allows Claude “to read, click, and navigate websites alongside you”. Example: filling in the same supplier form every week, with you approving before anything is sent.

Source: Claude help: Get started with Claude in Chrome

CLAUDE.md is a plain text file in your folder that Claude Code reads at the start of every session. Anthropic calls these files “instructions you write to give Claude persistent context”. For a business, it works like an onboarding sheet: what you sell, how you write, how you want Claude to work.

How to give Claude a memory of your businessSource: Claude Code memory

Permission modes decide how much Claude Code may do without asking you first. In Manual mode it asks before edits and most commands; other modes let it accept edits on its own, only plan, or work more independently. Start in Manual mode and give more freedom only for work you have seen go right many times.

How much can Claude Code do without asking?Source: Claude Code permission modes

A subagent is a helper Claude starts for one self-contained job, with its own separate memory, that reports back when done. Anthropic calls subagents “specialized AI assistants that handle specific types of tasks”. Useful for research or a review that would otherwise flood your main conversation; not needed on day one.

Source: Claude Code subagents

Hooks are small automatic actions that run at a set moment in Claude Code, for example every time Claude finishes a task. Use them when something must happen every time, without relying on Claude to remember it. Most business owners never need one.

Source: Claude Code hooks

Paid Claude plans have “a session limit that resets every five hours, and weekly limits that reset once every week”, in Anthropic's words. Chats and Claude Code count against the same limits, so a long build session uses the same allowance as your questions.

Read the full guideSource: Claude help: How do usage and length limits work?

MCP, the Model Context Protocol, is an open standard that lets AI tools like Claude connect to other apps and data. The project describes it as “an open-source standard for connecting AI applications to external systems”. You rarely set it up yourself: most connectors use MCP behind the scenes.

Claude plugins and connectorsSource: Model Context Protocol: Introduction

A connector links Claude to an app you use, like Gmail or Google Drive. Anthropic says connectors “let Claude access your apps and services, retrieve your data, and take actions within connected services”. Connect only the apps a task needs.

Read the full guideSource: Claude help: Use connectors

A plugin is a ready-made package that adds skills, connectors and helpers to Claude in one install. Anthropic says each plugin “bundles skills, connectors, and sub-agents into a single package”. Install only plugins from sources you trust.

Read the full guideSource: Claude help: Use plugins in Claude