Eight Products. One Model. Total Confusion.
Anthropic now ships Claude Chat, Claude Code, Claude Cowork, Claude Dispatch, Claude Channels, Computer Use, Claude Managed Agents, and the Claude Agent SDK.
If you are confused about which one to use, you are not alone. Each one solves a different problem, runs in a different environment, and costs different money. Some overlap. Some do not. And the naming does not help.
Here is the honest breakdown, fact-checked against the actual docs and real usage.
The Quick Reference
| Product | What It Does | Who It Is For | Where It Runs |
|---|---|---|---|
| Claude Chat | Conversational AI | Everyone | Browser, desktop app, mobile |
| Claude Code | Agentic coding in your terminal | Developers | Your machine (CLI + IDE) |
| Claude Cowork | File agent with computer use, plugins, and sub-agents | Knowledge workers | Sandboxed VM on your desktop |
| Claude Dispatch | Smart router from phone to desktop | Mobile users | Phone paired to desktop |
| Claude Channels | Messaging bot (Telegram, Discord) | Power users | VPS or local machine |
| Computer Use | Controls your actual screen | Automation | Your machine via Cowork |
| Managed Agents | Cloud-hosted autonomous agents via API | Businesses, SaaS builders | Anthropic's cloud |
| Agent SDK | Build custom agent loops | Developers building AI products | Your infrastructure |
Claude Chat
What it is: The original. Open claude.ai or the desktop app, type a message, get a response. Writing, research, brainstorming, analysis, image generation.
What it is good at: Conversation. Quick questions. Document analysis. Drafting. Anything you can express in a chat window.
What it cannot do: Execute code on your machine. Read your file system. Run terminal commands. It stays inside the chat window.
Cost: Free tier, Pro ($20/month), Max ($100-200/month).
Use it when: You need to think through a problem, draft content, analyze a document, or brainstorm. This is the starting point for most people.
Claude Code
What it is: An agentic coding tool that runs in your terminal (CLI), VS Code, JetBrains, or as a desktop app. It reads your entire codebase, understands the context, and makes changes across multiple files autonomously. It executes terminal commands, runs tests, manages Git, and builds features.
What it is good at: Software engineering. It can read your project, plan changes, edit files, run tests, commit code, and push to GitHub. It has auto memory (remembers your preferences between sessions), skills (plug-and-play instruction files), and subagent spawning for parallel work.
What it cannot do: Natively handle document workflows like spreadsheets or presentations. It is built for code, not knowledge work.
Cost: Included with Claude Pro/Max subscription, or via API usage.
Use it when: You are building software. If your task involves code, repositories, tests, or deployment, this is the tool.
Key features:
- ●CLAUDE.md project instructions (read every session)
- ●Auto memory (learns your preferences over time)
- ●Skills (.claude/skills/ for domain expertise)
- ●Hooks (custom shell commands triggered by events)
- ●Subagents (spawn parallel workers for complex tasks)
- ●MCP servers (connect to external tools like Playwright, databases, APIs)
- ●Scheduled tasks via desktop app
Claude Cowork
What it is: A desktop agent that runs in a sandboxed virtual machine on your machine. It has access to your local files, connectors, plugins, and your apps through computer use. It can break complex work into smaller tasks using sub-agents.
What it is good at: Multi-step knowledge work. Spreadsheet analysis with real formulas. PowerPoint generation. Report creation. Data extraction. File organization. PDF processing. It writes actual files to a designated folder on your disk. It can also work inside external platforms like Notion through connectors.
What it can also do that many people miss:
- ●Scheduled tasks (run on-demand or on a recurring cadence)
- ●Sub-agent coordination (breaks complex work into parallel subtasks)
- ●Plugin support (connect to external services)
- ●Computer Use integration (can control desktop applications)
Limitations: Your computer must stay awake and the Claude Desktop app must remain open for tasks to run. Sessions cannot be shared with others. Memory works within projects but does not persist across standalone sessions. Uses more of your usage allocation than regular chat because multi-step tasks are compute-intensive.
Cost: Included with Pro ($20/mo), Max ($100-200/mo), Team ($30/user/mo), and Enterprise.
Use it when: Your task involves documents, data, or file management. A non-technical team member analyzing sales data, creating reports, or organizing files would use Cowork. Also useful for anyone who needs scheduled recurring tasks without writing code.
Claude Dispatch
What it is: A feature inside Cowork that lets you send tasks from your phone to your desktop. Launched March 17, 2026 as a research preview. Your phone sends instructions, your desktop does the work, results sync back.
What it actually does: Dispatch is smarter than a simple remote control. When you assign a task, Claude figures out what kind of work is needed and routes it automatically. Development tasks run in Claude Code. Knowledge work runs in Cowork. You do not choose the mode, Claude does.
How it works: One persistent conversation thread connects your phone and desktop across time and location. You assign a task, go about your day, and come back to finished work. All processing happens locally on your desktop using your local files, connectors, and plugins.
Requirements: Claude Desktop app must be installed and running on your computer. Your computer must be awake. Claude mobile app on your phone.
Cost: Included with Claude Pro/Max subscription.
Use it when: You want to assign work from your phone while away from your desk. Start a research task from your commute, check results when you get home.
Claude Channels
What it is: Connects Claude Code to messaging platforms. Currently supports Telegram and Discord via MCP plugins. You text your AI from your phone, and it runs Claude Code on a server.
What it is good at: Always-on AI assistant. We run ours on a $7/month VPS. Send a resume to the Telegram bot, it scores it automatically. Ask it to check server health, it SSHs in and reports back. It is Claude Code with a phone number.
How it differs from Dispatch: Dispatch routes to Cowork or Code on your desktop (must be awake). Channels runs Claude Code on a remote server (always on, 24/7). Channels is for power users who want a persistent, headless AI agent accessible from anywhere.
Cost: Claude subscription + VPS hosting (optional but recommended for 24/7).
Use it when: You want to text your AI and have it actually do things on a server. Not just chat, but run commands, check systems, process files, 24 hours a day.
Computer Use
What it is: Claude controls your actual screen. It can see your desktop, move the mouse, click buttons, type into applications, and navigate GUIs. Available through Cowork since March 23, 2026.
What it is good at: Automating GUI-based workflows that have no API. Filling out forms in legacy software, navigating web applications, testing UI flows, interacting with desktop apps that only have a visual interface.
What it cannot do: Work faster than a human at clicking. It is deliberate and methodical. Best for tasks where accuracy matters more than speed.
Cost: Included with Claude subscription. Token-intensive due to screenshot processing.
Use it when: The task requires interacting with a GUI that has no API or CLI alternative. Legacy enterprise software, web apps without automation hooks, or testing UI flows.
Get the Weekly IT + AI Roundup
What changed this week in NinjaOne, ServiceNow, CrowdStrike, and AI. One email, every Monday.
No spam, unsubscribe anytime. Privacy Policy
Claude Managed Agents
What it is: Anthropic's newest product (April 2026). Cloud-hosted containers where Claude runs as an autonomous agent. You define the agent (model, prompt, tools, MCP servers), create an environment (container with packages and network rules), and start sessions via API. Anthropic handles all infrastructure.
What it is good at: Long-running async tasks. Research that takes hours. Batch processing. Building SaaS products that embed Claude as an agent. Notion has already integrated it for task delegation directly in their product.
What it cannot do yet: Schedule itself (no cron or heartbeat triggers). No native webhook triggers. Requires external orchestration (n8n, trigger.dev, or similar) to automate. Only supports remote HTTP MCP servers, not local ones.
Cost: Standard API token pricing + $0.08 per session-hour (active time only, idle is free) + $10 per 1,000 web searches.
Coming soon (research preview, apply for access):
- ●Outcomes: Agent self-evaluates against a rubric and iterates until criteria are met
- ●Multi-agent: Coordinator agent delegates to specialized callable agents (one level deep)
- ●Persistent memory: Knowledge stores that survive across sessions, with version history and audit trails
Use it when: You are building a product that needs Claude as a backend agent. SaaS integrations, enterprise workflows, research pipelines. This is an API product for developers, not an end-user tool.
Claude Agent SDK
What it is: The same tools, agent loop, and context management that power Claude Code, packaged as a Python and TypeScript SDK. You build your own agent from scratch with full control over every decision.
What it is good at: Custom agent applications where you need total control. You decide the tools, the flow, the error handling, the deployment. It is battle-tested because it is the same code that runs Claude Code internally at Anthropic.
What it cannot do: Give you a managed runtime. You handle deployment, containers, scaling, and monitoring yourself.
Cost: API token usage only. No infrastructure fees to Anthropic.
Use it when: You are a developer building a custom AI agent and you need full control over the agent loop, tool execution, and deployment. This is the power user option for people comfortable managing their own infrastructure.
How This Compares to Microsoft Copilot Studio
If Managed Agents reminds you of Microsoft Copilot Studio, it should. They solve the same problem from different ecosystems.
| Aspect | Claude Managed Agents | Microsoft Copilot Studio |
|---|---|---|
| What it is | Cloud agent builder via API | Cloud agent builder via low-code UI |
| Runtime | Anthropic cloud containers | Azure cloud |
| Model | Claude (Opus, Sonnet, Haiku) | GPT-4o, GPT-4.1, or Claude via Copilot Cowork |
| Integration | MCP servers, custom tools, any API | M365, Dataverse, Power Platform, Azure |
| Auth | Vault-based credential storage | Azure AD, enterprise SSO |
| Triggers | API call only (no cron or webhooks yet) | Webhooks, scheduled, event-driven, Teams triggers |
| Target user | Developers building SaaS products | Enterprise teams on Microsoft 365 |
| Pricing | $0.08/hr + tokens + $10/1K searches | Per-message pricing, varies by plan |
| Governance | Beta, evolving | Enterprise-grade, SOC2, GDPR, audit logs |
The connection most people miss: Microsoft and Anthropic collaborated on Copilot Cowork (Wave 3 of M365 Copilot). The technology that powers Claude Cowork also powers parts of Microsoft 365 Copilot Cowork. Same AI engine, different wrapper.
Bottom line: If your organization lives in Microsoft 365, Copilot Studio gives you deeper integration and enterprise governance. If you are building a standalone product or want model flexibility, Managed Agents gives you more control and a cleaner API. Same concept, different ecosystems.
The Decision Tree
"I just want to ask Claude a question."
Use Claude Chat.
"I need to build or modify software."
Use Claude Code.
"I need to process documents, spreadsheets, or organize files."
Use Claude Cowork.
"I want to send tasks from my phone while away from my desk."
Use Claude Dispatch (routes to Code or Cowork automatically).
"I want to text my AI 24/7 and have it run commands on a server."
Use Claude Channels on a VPS.
"I want Claude to control my screen and click through a GUI."
Use Computer Use (via Cowork).
"I am building a product and need Claude running in the cloud as a backend agent."
Use Managed Agents.
"I want to build a completely custom agent with my own infrastructure."
Use the Agent SDK.
"I want an always-on AI assistant that checks things on a schedule."
Use Claude Code on a VPS with Channels + cron, or Cowork's built-in scheduled tasks. Managed Agents cannot do this yet.
The Honest Comparison Table
| Feature | Chat | Code | Cowork | Dispatch | Channels | Computer Use | Managed Agents | Agent SDK |
|---|---|---|---|---|---|---|---|---|
| Conversational AI | Yes | Yes | Yes | Yes | Yes | No | Yes | Yes |
| File read/write | No | Yes | Yes | Via Cowork/Code | Via Code | Via GUI | Yes | You build |
| Terminal/bash | No | Yes | No | Via Code | Via Code | No | Yes | You build |
| Web search | Yes | Yes | Plugins | No | Yes | Via browser | Yes | You build |
| Code execution | No | Yes | Sandboxed | Routed | Via Code | No | Yes | You build |
| GUI control | No | No | Yes | Via Cowork | No | Yes | No | No |
| Memory | Projects | Auto memory | Projects | Shared | Auto memory | No | Memory stores | You build |
| MCP servers | No | Yes | Connectors | No | Yes | No | Remote HTTP only | You build |
| Scheduled tasks | No | Desktop app | Yes | No | Cron/watcher | No | No (biggest gap) | You build |
| Mobile access | App | No | No | Yes | Telegram/Discord | No | API only | You build |
| Multi-agent | No | Subagents | Sub-agents | No | Subagents | No | Callable agents | You build |
| Always-on | No | No | No (must be awake) | No | Yes (VPS) | No | Yes (cloud) | You build |
| Cost | Free-$200/mo | Subscription | Subscription | Subscription | Sub + VPS | Subscription | $0.08/hr + tokens | Tokens only |
The Bottom Line
Most people need exactly two of these products:
- ●Claude Chat for thinking and writing
- ●Claude Code if you write software, Cowork if you do not
Everything else is for specific workflows. Channels turns Claude into an always-on assistant. Dispatch lets you work from your phone. Computer Use automates GUIs. Managed Agents embeds Claude into your SaaS product. The Agent SDK gives you full control.
Do not try to learn all eight at once. Start with what you need today. Add tools as your workflow demands them.
*For a hands-on walkthrough of Claude's four automation modes, read 4 Ways Claude Works When You're Not at the Keyboard.*
*Want to find which AI model fits your workflow? Take the free quiz and get matched in 2 minutes.*