Anthropic offers Claude in multiple forms: chat, coding, long-running work, design, browser assistance, Microsoft 365 integration, managed agent infrastructure, and a developer SDK. The current product and feature names are grounded in Claude’s official site and developer documentation.[1][2][3][4]
Product Overview
graph TD
A[Claude Product Lineup] --> B[Claude Chat]
A --> C[Claude Code]
A --> D[Claude Cowork]
A --> E[Claude Design]
A --> F[Claude for Chrome]
A --> G[Claude for Microsoft 365]
A --> H[Claude Managed Agents]
A --> I[Claude Agent SDK]
B --> B1[General & Business Users]
C --> C1[Developers & Engineers]
D --> D1[Teams & Organizations]
E --> E1[Designers & Frontend Engineers]
F --> F1[Browser-based assistance]
G --> G1[Microsoft 365 integration]
H --> H1[Long-running async agents]
I --> I1[Custom Agent Development]Claude Chat
Claude Chat is Anthropic’s main chat interface, accessible at claude.com/chat. It runs entirely in the browser and handles a wide range of tasks: writing, coding assistance, data analysis, research, and learning — no installation required.
Key features:
- Projects — Upload files and maintain context across conversations. Reference technical specs, meeting notes, or code files as the basis for ongoing work.
- Artifacts — Generated code, documents, SVGs, and HTML render in a separate panel for preview and iteration. Changes to the artifact are reflected immediately without scrolling through the chat.
- Model selection — Claude Fable, Opus, Sonnet, Haiku, and other available models are selected based on task and cost requirements. For API use, confirm the current model list in the model documentation.
- Research and Memory — Research supports multi-step investigation, while Memory preserves preferences and work context across conversations.
- File uploads — Attach PDFs, images, spreadsheets, and other files for direct analysis and Q&A.
Target users: General users to business users. No programming knowledge required. The fastest way to start using Claude.
Claude Code
Claude Code is a terminal CLI tool for developers (claude command). It reads and understands the full structure of a codebase — files, dependencies, test results — and then acts: editing files, running tests, executing git commands, and calling shell tools. It functions as an autonomous agent rather than a simple code completion tool.
Core operations:
- Read, write, and edit files across a codebase (including multi-file changes)
- Run tests, check results, and fix failures in a single loop
- Execute git operations: commit, branch, diff, stash
- Run shell commands: npm, cargo, pytest, and others
- Connect to external tools via MCP server integrations
Core Concepts
The following concepts are central to using Claude Code effectively.
CLAUDE.md is a project-specific contract file that defines rules, context, and constraints for Claude Code. When Claude Code starts, it automatically reads CLAUDE.md from the repository root to understand the project scope, prohibited actions, available commands, and coding conventions. Centralizing team rules in this file customizes Claude’s behavior for the project.
Skill is a reusable instruction or knowledge definition that extends Claude Code. You write instructions in SKILL.md, and Claude can load the skill automatically when relevant or you can invoke it explicitly with /skill-name. In current Claude Code, custom commands have been merged into skills, while existing .claude/commands/ files continue to work.[5]
Agent is a sub-agent definition with a specific area of expertise, set of tools, and context. Roles such as “documentation agent” or “security review agent” can be defined separately and launched from the main Claude instance, running in parallel.
Sub Agent is a child agent spawned by an agent. Sub agents enable a swarm architecture where large tasks are divided and processed in parallel by multiple independent agents.
Rules are project-specific constraints governing safety, quality, and style. Stored in directories like shared/rules/ and referenced from CLAUDE.md, they enforce policies such as “never run npm run build without explicit approval” and reduce the risk of accidental destructive operations.
Hooks are scripts that run in response to Claude Code operation events — before and after file writes, before and after command execution, and so on. Hooks enable automatic test runs, change logging, and notification triggers without manual intervention.
settings.json is the Claude Code configuration file. It defines permitted shell commands, tools, environment variables, and MCP server connections. It serves as the enforcement layer for security policies that restrict Claude’s operational scope.
Command is the older custom slash-command format placed in /.claude/commands/. Skills are now the recommended format, but existing .claude/commands/ files still use the same invocation style.[5]
graph LR
CLAUDE_MD[CLAUDE.md] --> |read on startup| CC[Claude Code]
SKILL[Skill] --> |auto-loaded or /skill-name| CC
RULES[Rules] --> |referenced as constraints| CC
HOOKS[Hooks] --> |intercept events| CC
SETTINGS[settings.json] --> |permissions & tools| CC
CC --> |spawns| AGENT[Agent]
AGENT --> |divides and delegates| SUBAGENT[Sub Agent]Target users: Software engineers and developers comfortable working in a terminal.
Claude Cowork
Claude Cowork is a work agent for handing outcomes to Claude and steering long-running work across web, desktop, and mobile. Claude can continue tasks, organize files, research, and produce decks, spreadsheets, or documents for the user to review.[1]
Key features:
- Continued execution — Claude can keep working after the laptop is closed and can run scheduled tasks.
- Parallel work — Claude can split larger work into chunks and research, organize, and draft in parallel.
- Permission controls — Users choose folders and connected tools, and significant actions can require approval. Team and Enterprise admins can manage feature access and permissions.
Target users: Business teams, product teams, and organizations exploring AI adoption at scale.
Claude Design
Claude Design is an AI feature for UI design and prototype generation. Describe a design intent in text, and Claude Design produces implementable code or visual prototypes.
Key features:
- UI component code generation — Generates React and HTML/CSS components from text instructions. Converts design mockups into working code immediately.
- Design system integration — References existing design tokens (colors, typography, spacing) when generating components, maintaining consistency with established brand guidelines.
- Visual prototype output — Outputs are reviewable in a visual interface similar to Figma, enabling iterative refinement with designers and engineers working from a shared artifact.
- Accessibility support — Code generation considers ARIA labels, color contrast ratios, and keyboard navigation to produce accessible output by default.
Target users: UI designers, frontend engineers, and product managers who prototype interfaces.
Claude for Chrome
Claude for Chrome brings Claude into the Chrome browser so it can work with the page the user is viewing. Claude Code also has a separate computer use preview for browser operation. Use these capabilities for page-aware research, assisted form work, and checking browser-based workflows.[1][2]
Key features:
- Screenshot capture and analysis — Claude captures the current state of a page as a screenshot, interprets the content, and decides the next action accordingly.
- Form interaction — Automates input, submission, and confirmation steps in web forms. Suitable for repetitive data entry tasks that occur at regular intervals.
- Information gathering and web scraping — Collects information across multiple pages automatically. Applicable for research and aggregation of publicly available web content.
- Security considerations — Browser operations should always run in a sandboxed environment. Extra care is required when handling production systems or authentication credentials.
Use cases: Web scraping, automated form submission, UI behavior testing, and routine web research tasks.
Target users: Automation engineers, QA engineers, and users managing data collection workflows.
Claude for Microsoft 365
Claude for Microsoft 365 connects Claude with Microsoft 365 work context such as documents, email, and calendar information. It helps with meeting preparation, document creation, and information synthesis inside Microsoft 365-based workflows.[1]
Key features:
- Document, email, and calendar context — Claude can use Microsoft 365 context to summarize information and prepare work artifacts.
- Work-aware responses — It helps synthesize internal information for meetings, customer work, and knowledge tasks.
- Managed connection scope — Data access follows the organization’s connection and permission design.
Target users: Teams and enterprise users that use Microsoft 365 as their work platform.
Claude Managed Agents
Claude Managed Agents is Anthropic’s managed agent runtime. Instead of building your own agent loop, tool execution, sandboxing, and runtime, you can run long-running asynchronous tasks where Claude reads files, runs commands, searches or fetches the web, and connects through MCP. As of July 2026 it is in beta, and Managed Agents API calls require the managed-agents-2026-04-01 beta header.[3]
Key features:
- Agent / Environment / Session / Events — The model, system prompt, tools, MCP servers, and skills are defined as an Agent, while environments and sessions are managed separately.
- Cloud or self-hosted sandboxes — Run in an Anthropic-managed cloud sandbox or a self-hosted sandbox on your own infrastructure.
- Long-running async execution — Suited to tasks that run for minutes or hours, use multiple tools, keep stateful sessions, or run on a schedule.
- Built-in tools — Bash, file operations, web search and fetch, and MCP server connections are available.
Target users: Developers, AI engineers, and platform teams that need managed infrastructure for long-running asynchronous agent work.
Claude Agent SDK
Claude Agent SDK is the SDK for building custom AI agents programmatically. Available in Python and TypeScript, it covers agent definition, tool registration, execution management, and multi-agent orchestration.
Key features:
- Agent definition — Define system prompts, available tools, and model selection in code. Provides precise control over agent behavior.
- Tool registration — Register functions, API calls, and database queries as tools that the agent selects and calls autonomously during execution.
- Execution management — Manage agent state, logs, and errors programmatically. Handles lifecycle management for long-running agents.
- Multi-agent orchestration — Implement orchestrator–sub-agent patterns to coordinate multiple agents. Enables parallel and distributed processing of large tasks.
- Claude Code integration — Claude Code’s agent capabilities use this SDK internally. The SDK is also available for local development and testing of agent behavior.
graph TD
SDK[Claude Agent SDK] --> ORCH[Orchestrator Agent]
ORCH --> SA1[Sub Agent A]
ORCH --> SA2[Sub Agent B]
ORCH --> SA3[Sub Agent C]
SA1 --> T1[Tool: Web Search]
SA2 --> T2[Tool: Code Execution]
SA3 --> T3[Tool: Database]Target users: Software engineers, AI engineers, and developers integrating custom agents into production systems.
Choosing the Right Product
| Use case | Recommended product |
|---|---|
| Chat with an AI assistant | Claude Chat |
| Write and execute code from a terminal | Claude Code |
| Use AI across a team | Claude Cowork |
| Generate UI designs and prototypes | Claude Design |
| Work with browser pages | Claude for Chrome |
| Use Microsoft 365 work context | Claude for Microsoft 365 |
| Run long-running asynchronous agents | Claude Managed Agents |
| Build a custom agent in code | Claude Agent SDK |
FAQ
Q: What is the difference between Claude Chat and Claude Code?
Claude Chat is a browser-based chat interface usable without programming knowledge. Claude Code is a terminal CLI that operates directly on a codebase — editing files, running tests, and executing git commands autonomously. The key distinction is that Claude Code takes real actions on a local project rather than generating text to copy elsewhere.
Q: How do Claude Agent SDK and Claude Managed Agents differ?
Claude Agent SDK lets developers use the same agent loop and tools that power Claude Code from Python or TypeScript. Claude Managed Agents provides Anthropic-managed infrastructure and session handling for long-running agent work. Choose the SDK when you need programmatic application control; choose Managed Agents when the managed runtime and asynchronous execution model are the main requirement.[3][4]
Q: Is CLAUDE.md required for every project?
It is not required, but defining project-specific rules — prohibited commands, coding conventions, directory structure — in CLAUDE.md optimizes Claude Code’s behavior for the project. It is particularly effective in team development environments.
Q: What are the security risks when using Claude for Chrome or computer use?
When Claude works with web pages or browser operations, sites that handle authentication or sensitive data require careful scope control. Limit access, require approval for important actions, and keep execution logs reviewable.
Detailed Guides in This Section
- Claude Model Comparison & Selection Guide — Opus, Sonnet, and Haiku characteristics with use-case-based selection criteria
- Claude API Basics and Prompt Caching — SDK fundamentals, cost structure, and caching strategies
- Claude Code × MCP Integration Guide — Configuring MCP servers in settings.json and usage patterns
- Multi-Agent Design Patterns — Orchestrator-Worker, Pipeline, Specialist Pool, and Swarm patterns
- Anthropic’s Safety Philosophy and Claude’s Design — Constitutional AI, HHH principles, and RSP explained
- Claude Managed Agents: Getting Started — managed agent runtime, use cases, and comparison with other agent tools
This page uses Claude’s official site for the product lineup, Claude Code documentation for execution surfaces and Skills behavior, and the Managed Agents and Agent SDK documentation for developer-facing agent runtime claims.[1][2][3][4][5]
References
- Anthropic, Claude
- Anthropic, Claude Code overview
- Anthropic, Claude Managed Agents overview
- Anthropic, Agent SDK overview
- Anthropic, Extend Claude with skills
For the latest releases and updates, check the official website and official documentation.