The Claude landscape

You are here · All paths · Free plan for the exercises · 30–40 min · Assumes What Claude actually is.

What you’ll learn: every place Claude exists, what each one is for, and how they relate — so you always know which door to walk through.

If you only read one thing

Claude isn’t one product, it’s about a dozen, and they all run on the same brain. What differs is what each one is allowed to touch.

If you’re not a developer, you need to know four names:

• claude.ai / the desktop app — the chat window. Asking, writing, analysing.
• Projects — a chat window that already knows your background material, so you stop re-explaining yourself.
• Cowork — you hand Claude a folder and a job, walk away, and come back to finished work.
• Connectors — plug-ins that let Claude see your email, calendar, Drive, Slack and so on.

The rest of this module maps everything else, including the developer tools. Skim those sections; you’re reading them so the names aren’t strange later, not to learn them now.

The map

Underneath everything is one set of models — Fable 5, Opus 5, Sonnet 5, Haiku 4.5. On top of them sit four layers of product. They all think the same; they differ in what they’re allowed to touch.

Most people live in layers 1 and 2 and never need 3 or 4. If you’re on Path A, that’s the whole map — layers 3 and 4 are there so you know they exist.

And cutting across all of them, four extension mechanisms you’ll meet repeatedly:

• Skills — packaged instructions and workflows Claude loads on demand
• MCP / Connectors — a standard protocol for plugging Claude into external systems
• Plugins — bundles of skills, agents, hooks, and MCP servers, distributed via marketplaces
• Memory — persistent context across sessions

Layer 1 — Chat

claude.ai (web), desktop app, mobile app

The conversational interface. Same account, same conversations, synced across all three.

What lives here:

Use it for: thinking, writing, analysis, one-off questions, anything conversational.

Layer 2 — Agentic desktop

Both of these give Claude a loop: it plans, acts, observes the result, and continues — without you approving every step.

Claude Cowork

The non-developer agent. A visual interface where Claude works on multi-step tasks against your local files — research synthesis, file organisation, document generation — with visible progress and the ability to steer mid-task.

• Available on all paid plans (Pro, Max, Team, Enterprise)
• Full experience on the Claude Desktop app (macOS, Windows, Linux beta), where Claude can also use your local files and browser
• Also available on web and mobile for handing off and checking on tasks
• Supports Skills, Connectors, Plugins, scheduled tasks, and computer use

Use it for: “here is a folder, do a job with it” — no terminal, no code.

Covered in Cowork.

Claude Code

The developer agent. Reads your codebase, edits files, runs commands, and integrates with your dev tools.

Surfaces: terminal CLI, VS Code, JetBrains, desktop app, web (claude.ai/code), mobile, Slack, GitHub Actions, GitLab CI. All surfaces share the same engine — your CLAUDE.md, settings, and MCP servers work across every one of them.

Use it for: anything involving a repository.

Covered in all of stage 03.

Cowork vs Claude Code: same underlying agentic engine, different audience. Cowork is file-and-document oriented with a GUI; Claude Code is repository oriented and code-native. If your task has a git directory in it, use Claude Code.

Layer 3 — In-app integrations

Claude showing up where you already work.

Covered in Claude inside your apps.

Layer 4 — Developer platform

Claude API

Direct model access over HTTP. The Messages API is the core endpoint. Official SDKs in Python, TypeScript, Go, Java, C#, PHP, Ruby, plus an OpenAI-compatibility shim and a first-party CLI (ant).

Also available through Amazon Bedrock, Claude Platform on AWS, Google Cloud, and Microsoft Foundry.

Covered in stage 04.

Claude Agent SDK

Claude Code as a library. You get the agentic loop, the tool suite, hooks, permissions, subagents, skills, and session management — with full control over orchestration. Python and TypeScript.

Covered in The Claude Agent SDK.

Managed Agents

Anthropic-hosted agents that run in a managed cloud sandbox. You define the agent; Anthropic runs it. Supports scheduled deployments, webhooks, memory stores, and multi-agent orchestration without you operating infrastructure.

Covered in Managed Agents.

Claude Console

The web dashboard at platform.claude.com: API keys, workspaces, spend limits, usage analytics, the prompt workbench, and evaluation tooling.

The cross-cutting mechanisms

These four appear at every layer. Learning them once pays off everywhere.

Skills

A SKILL.md file with YAML frontmatter (name, description) plus optional bundled files and scripts. Claude loads the description always, the body when triggered, and bundled resources only when needed — progressive disclosure. Roughly 100 tokens per installed skill until it fires.

Works in: claude.ai (upload as zip), Cowork, Claude Code (~/.claude/skills/ or .claude/skills/), the API (skill_id + code execution tool), Agent SDK, Managed Agents.

Skills do not sync between surfaces. A skill uploaded to claude.ai is not available in the API or in Claude Code. Upload separately per surface.

MCP (Model Context Protocol)

An open standard for connecting AI tools to external data and actions. In consumer products these are called Connectors. In Claude Code and the SDK they’re MCP servers. Same protocol.

Plugins

Installable bundles that can contain skills, subagents, hooks, MCP servers, and commands. Distributed through marketplaces. Available in Claude Code and in Cowork.

Memory

• claude.ai — Claude remembers preferences and past context across chats; importable and exportable
• Claude Code — CLAUDE.md files you write, plus auto memory Claude writes for itself per repository
• API / Managed Agents — the memory tool and memory stores

Choosing a door: a decision table

Try it

Exercise 1. Open claude.ai and find, in the UI: the model picker, the attachment button, the search/research toggles, and Settings → Features. Just locate them.

Exercise 2. Install the Claude Desktop app from claude.com/download. Note the three tabs: Chat, Cowork, Code.

Exercise 3. Take the three most repetitive things you do at work and place each one in the decision table above. Write the answers in your prompt journal. You’ll revisit this at the end of stage 02.

Checkpoint

Without looking, name: three surfaces where Skills work, the difference between a Connector and an MCP server, and when you’d pick Cowork over Claude Code.

Going deeper

• Where can I access Claude?
• Claude Code overview
• Get started with Claude Cowork
• Agent Skills overview
• Browse skills, connectors, and plugins in one directory