Claudeは実は4つの異なる製品である

Most people say "I use Claude" like it's one thing. It's not. Claude is four distinct products with four completely different jobs. Use the wrong one and you're either overcomplicating simple tasks or leaving massive productivity on the table. Here's what each one actually does, who it's for, and when to pick one over the others. PS: I built a Google Doc that you can hand to Claude Code to deploy your first Claude Managed Agent (Claude's newest product) in 60 seconds. This is the easiest way to test this new feature if you're non-technical. Grab it free here. -------------------------------------- Claude Chat: The Starting Point This is the Claude most people know. Go to claude.ai, type a question, get an answer. What it does: General conversation and Q&A Writing, editing, and brainstorming Document analysis (upload PDFs, images, spreadsheets) Research and summarization Projects for saving reusable context and instructions Who it's for: Anyone who needs a smart assistant for thinking work. Writers, researchers, students, business owners who want quick answers. If your work is mostly "help me think through this" or "help me write this," Claude Chat is your tool. Where it works: Web browser, desktop app, mobile app. No setup. Just start typing. What it can't do: This is the critical limitation. Claude Chat can't take action in your other tools. It can't send an email, organize your files, update a spreadsheet, or commit code. You're the middleman. You ask, it answers, you copy-paste the result somewhere else. Pricing: Free tier available. Pro is $20/month for significantly more usage. Bottom line: Claude Chat is your thinking partner. Great for brainstorming, writing, and analysis. But it stays inside the chat window. Claude Code: The Developer's Workhorse Claude Code is Claude with hands, specifically for software development. What it does: Reads your entire codebase and understands project context Edits files directly in your repository Runs terminal commands Handles git workflows (commits, PRs, branches) Spawns Agent Teams where multiple Claude instances coordinate on complex tasks Integrates with GitHub, GitLab, Slack via MCP connectors Who it's for: Developers. Full stop. If you don't write code, Claude Code isn't for you. But if you do write code (any language, any framework), this is the single biggest productivity upgrade available right now. It lives in your terminal, your IDE (VS Code, Cursor, etc.), or your browser. It doesn't just suggest code. It writes it, tests it, and commits it. Where it works: Terminal, VS Code, Cursor IDEs, browser-based environments. What it can't do: Claude Code is purpose-built for software development. It won't organize your Google Drive, draft your emails, or manage your calendar. It's also not beginner-friendly. If you've never opened a terminal, the learning curve is real. Pricing: Same tiers as Claude Chat. Pro at $20/month, up to Ultra at $200/month for heavy usage. Bottom line: If you write code and you're not using Claude Code, you're working harder than you need to. It's a junior developer that never sleeps, never complains, and ships code 24/7. Claude Cowork: The Biggest Unlock for Non-Developers This is the one most people haven't tried yet. And it's the one that matters most for non-developers. Claude Cowork lets Claude take actions directly on your computer and inside your apps. Not just answer questions. Actually do the work. As of April 2026, Cowork is now generally available on both macOS and Windows for all paid subscribers. It's no longer a research preview. What it does: Connects to Google Drive, Gmail, Excel, PowerPoint, Slack, Zoom, and more Organizes files, renames documents, deletes duplicates Generates reports, spreadsheets, and presentations Runs batch operations (convert formats, compress images, standardize naming) Automates browser tasks via Chrome extension Uses customizable plugins for specific workflows Who it's for: Knowledge workers. Solo operators. Small business owners. Agency teams. Anyone drowning in repetitive computer tasks who doesn't write code. If you spend your day bouncing between Google Docs, email, spreadsheets, and Slack, Cowork is built for exactly your workflow. Where it works: Claude Desktop app (macOS and Windows). Requires the app to stay open during tasks. What it can't do: Cowork uses more tokens than regular chat, so plan your usage. It needs the desktop app running. And there's a learning curve to understanding what it can and can't do across your specific tool stack. Pricing: Available on any paid Claude plan (starting at $20/month). Bottom line: Cowork is Claude living inside your business tools. It pulls data from one app, processes it, and pushes results to another. No code required. Claude Managed Agents: The Platform Play This is the newest Claude product and the most different from the other three. Managed Agents launched in public beta in April 2026. It's not a chat interface or a desktop tool. It's infrastructure. Claude Managed Agents is a hosted platform for building and deploying autonomous AI agents at scale. You define the agent, give it tools, and let it run in the cloud for minutes or hours without supervision. What it does: Runs Claude as a long-running autonomous agent in cloud infrastructure Provides a pre-built agent harness (the loop that calls Claude and routes tool calls) Gives each agent a secure sandbox with bash, file operations, web search, and code execution Supports MCP servers so agents can connect to external services Handles session persistence, so an agent can pick up where it left off after interruption Scales to multiple agents coordinating on complex tasks (multi-agent workflows) Built-in prompt caching, context compaction, and performance optimizations Who it's for: Developers and businesses building AI-powered products. If you want to deploy Claude as a backend service that handles customer requests, processes documents, manages workflows, or automates operations without human involvement, this is the tool. Companies like Notion, Asana, and Sentry are already using it. It's designed for teams that want to embed Claude's intelligence into their own products, not just use it as a personal assistant. To quickly spin up your first Managed Agent you can hand this Google doc to Claude Code and simply tell it "deploy a managed agent" and it'll take it from there. Where it works: Cloud-hosted via the Claude Platform API. You interact with it through API calls, not a chat window or desktop app. What it can't do: This is not a consumer product. You won't find Managed Agents in the Claude app. It requires API access, technical implementation, and an understanding of how to design agent workflows. It's not for someone who just wants Claude to help write an email. It also consumes API credits, and long-running agent sessions can add up. Plan your usage and set spend limits. Pricing: Usage-based via the Claude API. You pay for the tokens your agents consume. Available to all API accounts. The architecture (simplified): Anthropic decoupled three components that used to be bundled together: The brain (Claude and its harness): Handles reasoning and deciding what to do next The hands (sandboxes and tools): Where Claude actually executes work The session (event log): A durable record of everything the agent did Each piece can fail and recover independently. If a sandbox crashes, the brain catches it and spins up a new one. If the harness fails, a new one loads the session log and picks up where it stopped. No single failure kills the whole agent. This separation also means the brain doesn't hold credentials. OAuth tokens and API keys live in a secure vault outside the sandbox, so even if something goes wrong in the execution environment, your sensitive data stays protected. Bottom line: Managed Agents is Claude as a service you deploy, not a tool you use personally. It's for building products powered by Claude that run autonomously in the cloud. The Comparison Table How to Pick the Right One Start with Claude Chat if: You're new to AI assistants Your work is mostly thinking, writing, or research You don't need Claude to take action anywhere You want the simplest possible experience Move to Claude Code if: You write code in any language You want AI that actually ships changes, not just suggests them You're comfortable in a terminal or IDE You need automated code review, testing, or deployment Move to Claude Cowork if: You need Claude to work inside your existing apps Your day involves repetitive tasks across multiple tools You're non-technical but want real automation You want to stop being the copy-paste middleman between Claude and everything else Build with Managed Agents if: You're building a product that needs AI agents running in the background You want Claude handling customer-facing workflows autonomously You need agents that run for hours, recover from failures, and scale You're embedding Claude's intelligence into your own software The Real Power Move These four tools aren't competitors. They're layers of the same platform. Use Claude Chat for thinking through problems and drafting content. Use Claude Code for building software. Use Cowork for the repetitive tasks that eat your day across your app stack. Use Managed Agents to deploy Claude-powered agents as a product or service. The progression for most people looks like this: start with Chat, realize you need it to take action, move to Cowork (or Code if you're a developer), and eventually build with Managed Agents when you want to scale beyond personal use. The question isn't "which Claude is best?" It's "which Claude fits this specific task?" Pick one, get comfortable, then expand when you hit its limits. You'll know exactly when you need the next one because you'll feel the friction. PS: I built a Google Doc that you can hand to Claude Code to deploy your first Claude Managed Agent (Claude's newest product) in 60 seconds. This is the easiest way to test this new feature if you're non-technical. Grab it free here.