squad

Claude Agents Compared: Cowork, Claude Code, Agent Teams, and Managed Agents

Claude agents means six different Anthropic products. This page compares Cowork, Claude Code, subagents, agent teams, Managed Agents, and the Agent SDK on price, limits, and fit.

Bhanu Teja Pachipulusu

Bhanu Teja Pachipulusu

· Updated November 4, 2026

Claude Agents, Compared which one fits the job

squadAI teammates to do your work

Claude agents is not one product. Anthropic ships six: Claude Cowork, Claude Code, Claude Code subagents, Claude Code agent teams, Claude Managed Agents, and the Claude Agent SDK. This page compares them as of November 4, 2026.

7x

more tokens than a standard session when Claude Code agent team teammates run in plan mode, according to Anthropic's cost documentation.

Source: Anthropic, Claude Code docs: Manage costs

iShort answer

Use Cowork for non-coding tasks you describe as a goal. Use Claude Code for software work, with subagents for side tasks. Treat agent teams as an experiment. Use Managed Agents or the Agent SDK only if you are building an agent into a product. None of the six is a task board for several specialist roles. Squad (formerly MissionControlHQ) fills that gap. It runs a squad, which is a group of AI teammates that each hold their own job, where most tools give you one agent.

Key takeaways

QuestionShort answer
What does Claude agents mean?Six Anthropic products: Cowork, Claude Code, subagents, agent teams, Managed Agents, and the Agent SDK
Cowork or Claude Code?Claude Code for work in a code repository. Cowork for documents, files, and research. Pro at $20 a month includes both
Subagents or agent teams?Subagents for focused side tasks. Agent teams are experimental, disabled by default, and cost about 7x the tokens in plan mode
What do Managed Agents cost?Standard token rates plus $0.08 per session-hour of running time. Beta, API only
Which Claude plan?Pro at $20 for light use, Max at $100 or Ultra at $200 for daily agent work. Managed Agents and the SDK bill through the API instead
What is missing for a business?A shared task board, schedules, per-role access, approvals, and a dashboard across several agents
Six Claude agents, three groups

Two products you use, two ways to run several agents inside Claude Code, and two ways to build your own.

1

Use it as a person

Claude Cowork for non-coding goals. Claude Code for software work. Both are included in paid Claude plans.

2

Run several agents in Claude Code

Subagents report back to one main agent inside a session. Agent teams are separate instances that message each other. Teams are experimental.

3

Build your own agent

Managed Agents is hosted by Anthropic and billed per token plus session-hour. The Agent SDK runs in your own process.

4

Run several business roles

None of the six is built as a standing task board for specialist roles. That layer is separate.

What Claude agents means

Claude agents means any of six Anthropic products that let Claude work toward a goal by using tools. They are built for different people and run in different places. Picking the wrong one is the most common mistake, because the names overlap.

Here is the short version of each.

The first four are included in a Claude subscription. The last two are for developers and bill through the Claude API.

Claude agents compared in one table

The table below compares all six on who each is for, where it runs, what it costs, and its main limit. Prices come from Anthropic's pricing pages on November 4, 2026.

OptionWho it is forWhere it runsPriceMain limit
Claude CoworkNon-coders and knowledge workersClaude app on desktop, web, and mobile; sessions and scheduled tasks run in the cloudIncluded in Pro ($20/mo), Max ($100/mo), Ultra ($200/mo), Team, EnterpriseSessions cannot be shared; uses limits faster than chat
Claude CodeDevelopersTerminal, IDE, desktop app, webIncluded in Pro and above; not on FreeShares plan usage limits; built for software work
SubagentsClaude Code usersInside one Claude Code sessionNo extra fee; counts toward the same plan usageOne session; results return to the main agent
Agent teamsDevelopers who want parallel agentsLocally, inside the lead's Claude Code sessionNo extra fee; about 7x tokens in plan modeExperimental; one team per session
Managed AgentsDevelopers building agent productsAnthropic-managed cloud sandbox, or a self-hosted sandboxAPI tokens plus $0.08 per session-hourBeta; you build the app around it
Agent SDKDevelopers building agents in their own codeA process you operateModel use on an API key; no SDK fee listed in the overviewYou build and host everything else

Claude Cowork

Claude Cowork product page showing the headline The work behind your best work and a notice that Claude Cowork is now just Claude
The Claude Cowork product page on October 30, 2026, with Anthropic's notice that Cowork is now just Claude.

Claude Cowork is Anthropic's agent for non-coding work. Anthropic's product page says you give it a goal, it works across your files and tools, and you come back to finished work for review. It uses the same underlying design as Claude Code, without a terminal.

Cowork runs on the Claude desktop app, on the web, and on mobile. Anthropic's support page says sessions run in the cloud on Anthropic's servers, so work continues when you close your laptop. Scheduled tasks run in the cloud with no device online, and Anthropic's page says they can run daily, weekly, or monthly. Cowork is available on paid plans only: Pro, Max, Ultra, Team, and Enterprise.

Three limits matter. Sessions cannot be shared with other people, though you can share individual artifacts. Cowork uses limits faster than chat, and Anthropic says so on its product page. Computer use inside Cowork is in research preview.

Cowork is also changing shape. On September 16, 2026, Anthropic announced that Cowork and chat are merging into one Claude, rolling out to Pro, Max, and Ultra first, with Team and Free to follow. An October 23 update says the merged experience is generally available on Claude Enterprise. The announcement says existing Cowork chats, projects, connectors, and skills carry over. The Cowork product page now opens with "Claude Cowork is now just Claude." In the merged app, you pick Cowork or Chat from the same message box.

For a longer look at how Cowork fits with a standing team of agents, see mission control for Claude Cowork.

Claude Code

Claude Code is Anthropic's coding agent. Anthropic's documentation says it reads your codebase, edits files, runs commands, and connects to development tools. It runs in a terminal, in VS Code and JetBrains, in a desktop app, and on the web and mobile.

Claude Code is included in Pro, Max, Ultra, Team, and Enterprise. The Free plan does not include it. You can also use it with an Anthropic Console account billed per token.

It has several ways to run work without you at the keyboard. Cloud scheduled tasks, which Anthropic used to call routines, run in the cloud with your computer off. They can start on a schedule, an API call, or a GitHub event. Desktop scheduled tasks run on your machine, and /loop repeats a prompt inside a session.

Claude Code also has the parallel options covered below: subagents, agent teams, and background agents that run several full sessions on one screen. For how a hosted squad sits next to it, see mission control for Claude Code.

Claude Cowork vs Claude Code

Claude Code is the better choice when the work lives in a code repository, and Cowork is the better choice for everything else. The deciding question is where the result ends up. If it ends up as a commit or a pull request, use Claude Code. If it ends up as a document, a cleaned folder, a spreadsheet, or a research summary, use Cowork.

The two share one engine. Anthropic's support page says Cowork uses the same underlying architecture as Claude Code and does not need a terminal. A person who writes code and also drafts reports can use both under one paid plan.

QuestionClaude CoworkClaude Code
Main jobDocuments, files, research, recurring personal tasksBuilding features, fixing bugs, commits, pull requests
InterfaceClaude app on desktop, web, mobileTerminal, IDE, desktop app, web
Needs a terminalNoNo, but the CLI is the full-featured surface
Scheduled workYes, runs in the cloudYes, cloud and desktop scheduled tasks
Parallel agentsPlugins can bundle sub-agentsSubagents, agent teams, background agents
PlansPro, Max, Ultra, Team, EnterprisePro, Max, Ultra, Team, Enterprise
Sharing sessionsNot possibleNot a listed feature

Winner for coding: Claude Code, because it is built around repositories, git, tests, and IDEs. Winner for non-coding work: Cowork, because it needs no terminal and works on files, connectors, and a schedule.

The usage pool is a small catch. Anthropic's usage page says claude.ai, Claude Code, and Claude Desktop count toward the same limit. For Team and Enterprise seats, Anthropic's cost documentation says the allowance is shared with chat and Cowork. The usage page does not say how Cowork counts on Pro, Max, and Ultra.

Claude Code subagents

Subagents are helper agents that Claude Code starts inside one session. Each has its own context window, system prompt, tool access, and permissions. Anthropic says a subagent keeps side work, such as searches and logs, out of your main conversation and returns only a summary.

You define a subagent as a Markdown file with YAML frontmatter in .claude/agents/ or ~/.claude/agents/. Only the name and description are required. A subagent can run on sonnet, opus, haiku, fable, a full model ID, or inherit.

Anthropic lists these limits:

Subagents work inside a single session. When the session ends, they end with it.

Claude Code agent teams

Agent teams let one Claude Code session act as a lead and coordinate other Claude Code instances called teammates. Each teammate has its own context window. Teammates message each other directly, and they share a task list where they claim work.

Anthropic Claude Code documentation page titled Orchestrate teams of Claude Code sessions, with a warning that agent teams are experimental
Anthropic's agent teams documentation on October 30, 2026 labels the feature experimental and disabled by default.

Anthropic's documentation says agent teams are "experimental and disabled by default." You turn them on with this environment variable:

CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1

Teams do not start in non-interactive mode.

The documentation lists these limits:

Teams and tasks are stored locally under the session. Claude Code removes the team config when the session ends, and the task list stays on disk. Anthropic suggests starting with 3-5 teammates and with research or review tasks, where teammates do not edit the same files.

Agent teams vs subagents

Use subagents when the work is a focused side task and only the result matters. Use agent teams when several agents need to share findings, challenge each other, and coordinate on their own. Anthropic's documentation makes the same split.

SubagentsAgent teams
CommunicationReturn a result to the caller; subagents that Claude named can also message each otherTeammates message each other directly
CoordinationThe main agent manages the workShared task list and messages
ContextOwn window; results come back to the callerOwn window; fully independent
Token costLowerHigher; about 7x in plan mode
StatusDocumented featureExperimental, disabled by default
LifespanOne sessionOne session

Winner for most work: subagents, because they cost less and are not experimental. Winner for competing-hypothesis debugging and parallel review: agent teams, because Anthropic's examples show teammates testing different theories and arguing with each other.

Anthropic also warns that turning agent teams on changes ordinary delegation. While teams are enabled, a subagent that Claude names launches as a teammate.

Claude Managed Agents

Claude Managed Agents is a hosted agent service on the Claude API. Anthropic describes it as a pre-built, configurable agent harness that runs in managed infrastructure. Instead of writing your own agent loop and sandbox, you create an agent, an environment, and a session, and send events to it.

Claude Managed Agents overview in the Claude Platform docs, showing beta status and the managed-agents-2026-04-01 beta header
The Managed Agents overview on October 30, 2026 shows beta status and the required beta header.

The four concepts are agent, environment, session, and events. An agent is the model, system prompt, tools, MCP servers, and skills. An environment is where sessions run, either an Anthropic-managed cloud sandbox or a self-hosted sandbox on your own infrastructure. A session is a running agent in an environment.

Built-in tools include bash, file operations, web search and fetch, and MCP servers. Sessions can run for minutes or hours, keep their state, and resume after pauses.

Managed Agents is in beta on every model, and every request needs the managed-agents-2026-04-01 beta header. It is available on the Claude API, Claude Platform on AWS, and Claude Models on Google Cloud (Anthropic managed). Access needs a Claude API key from the Claude Console. Anthropic says sessions are stateful, so Managed Agents is not currently eligible for Zero Data Retention or HIPAA coverage.

Two newer pieces matter for scale. Scheduled deployments run an agent on a cron schedule, with up to 1,000 deployments per organization. Multiagent orchestration, also in beta, lets an agent delegate to subagents one level deep, with at most 25 child threads at a time. It also offers dynamic workflows, where the agent writes a program that runs many agents in phases and combines their results.

Claude Managed Agents pricing and what you still build

Managed Agents bills on two lines: tokens and session runtime. Tokens are billed at the standard rates on Anthropic's model pricing page. Session runtime is $0.08 per session-hour, and it counts only while the session status is running. Idle time, rescheduling, and terminated time do not count.

Web search inside a session costs $10 per 1,000 searches. The Batch API discount does not apply. Managed Agents endpoints are rate limited per organization to 300 create requests and 1,200 read requests per minute.

Runtime on its own is small. A session that ran for 8 hours a day over 30 days would add 240 hours, or $19.20, in runtime. The token bill is the larger number, and it depends on the model you pick.

Managed Agents provides the agent engine. It does not provide the product around it. The overview lists the agent, environment, session, and events. You still build:

Anthropic's Console offers a way to build in the browser, and the Console has a form for deployments. The overview lists no business task board, shared inbox, or approval queue.

Claude Agent SDK

The Claude Agent SDK is a library that gives you Claude Code's tools, agent loop, and context management in Python and TypeScript. Anthropic's overview says to use it when you want to embed Claude Code's agent in your own application, in a process you operate.

It includes built-in tools, hooks, subagents, MCP, permissions, sessions, skills, and plugins. You write the application. You host it. You pay for model use through an API key.

Anthropic adds one rule for products. Its overview says that, unless previously approved, Anthropic does not allow third party developers to offer claude.ai login or rate limits for their products, including agents built on the Agent SDK. It tells developers to use API key authentication.

The difference between the SDK and Managed Agents is who runs the agent. With the SDK, you run it. With Managed Agents, Anthropic hosts the agent loop and sandbox.

Which Claude plan you need

Pro at $20 a month covers Claude Code and Cowork for light use. Max at $100 a month and Ultra at $200 a month are for daily agent work. Developers building on Managed Agents or the SDK need an API account instead, because those products bill through the Claude API.

PlanPriceClaude CodeCoworkUsage
Free$0NoNoLimited
Pro$20/mo, or $17/mo billed annuallyYesYesBaseline, every 5 hours and every week
Max$100/moYesYes5x Pro, every 5 hours and every week
Ultra$200/moYesYes10x Pro every week; no 5-hour limit on verified accounts
Team Standard$25/seat monthly, $20/seat annualYesYesMore than Pro
Team Premium$125/seat monthly, $100/seat annualYesYes5x Standard
Enterprise$20/seat/mo billed annually, 20-seat minimumYesYesUsage billed at API rates
API (Console)Per tokenYes, by Console accountNoManaged Agents and SDK

Team plans cover 2 to 150 seats. Anthropic's pricing page also lists monthly API credits for building on the Claude Platform: $100 a month with Max, $200 a month with Ultra, and up to $500 a month pooled on Team. Anthropic says these API credits do not apply to usage in Claude, Claude Code, or Cowork, and they are separate from the usage credits described in the next section.

For a comparison of what a flat subscription and an API key each cost for agent work, see ChatGPT subscription vs API for agents. For the wider cost picture, see AI agent running costs.

Claude usage limits for agents

Claude plans limit use in two windows. Anthropic's usage page says paid plans have a session limit that resets every five hours and weekly limits that reset once a week. Verified Ultra accounts have no five-hour limit. Conversation length, features used, model, and effort level all change how fast you reach them.

Agents use limits faster than chat. Anthropic's Cowork page says Cowork consumes limits faster than chat. Its Claude Code documentation says subagent requests count toward the same limits as the main conversation, and that agent teams use about 7x the tokens of a standard session when teammates run in plan mode. Anthropic recommends Sonnet for teammates and small teams to control cost.

When you reach a limit, usage credits let you keep working. Anthropic's support page says Pro, Max, and Ultra subscribers can turn them on, and usage beyond the plan is billed at standard API rates. Credits are prepaid, with an optional monthly cap and optional auto-reload. The page says credits apply to Claude conversations and Claude Code. It does not say whether they apply to Cowork.

For a sense of API-billed cost, Anthropic's cost documentation says enterprise deployments average about $13 per developer per active day and $150-250 per developer per month. Costs stay under $30 per active day for 90% of users.

Managed Agents and the SDK sit outside the plan limits. They bill per token on an API account, plus session runtime for Managed Agents.

How to choose a Claude agent by job

Start with who does the work and where the result lands. The decision tree below covers the common cases.

What are you building or doing?

  • If software work in a code repository→Claude Code, with subagents for side tasks
  • If documents, files, research, or recurring personal tasks→Claude Cowork
  • If an agent inside your own product→Managed Agents or the Agent SDK
  • If several business roles that run on schedules→A layer above these tools, such as a hosted squad

If you build your own agent, who should run it?

  • If Anthropic hosts the agent loop and sandbox→Claude Managed Agents (beta)
  • If you run it in your own process→Claude Agent SDK

If you use Claude Code, do you need parallel agents?

  • If focused side tasks where only the result matters→Subagents
  • If agents that must debate or share findings, and you accept experimental→Agent teams
ScenarioBest pickWhy
Fix a bug and open a pull requestClaude CodeBuilt for repositories, git, and tests.
Turn a messy folder into a formatted reportClaude CoworkWorks on files and tools with no terminal.
Search a large codebase without flooding the chatSubagentsOwn context window; only a summary returns.
Review a pull request from security, speed, and test anglesAgent teamsTeammates review in parallel and share findings. Experimental.
Run a long agent task in a hosted sandbox from your appManaged AgentsAnthropic hosts the loop and sandbox. Beta.
Embed a Claude Code style agent in your own serviceAgent SDKA library you run in your own process.
A weekly personal report from your own filesCowork scheduled tasksRuns in the cloud with no device online.
Support, content, and finance follow-ups running every dayA hosted squadNeeds standing owners, schedules, and shared visibility.

What Claude agents do not cover for a business

A business that runs several specialist roles needs more than one capable agent. It needs a named owner per role, schedules, separate access per role, a place for approvals, and one view of what every role did. The six Claude products each cover a piece.

The pattern is the same across all six. Each gives you an agent or a way to run a few agents for a while. None is listed as a standing team with a shared board that keeps working for months.

That is a fair design. Anthropic builds the agent engine, and customers build around it. The question for a business owner is who builds the layer around it. Hiring a developer to build it on Managed Agents or the SDK is one answer. Buying it hosted is another. Teams that prefer to self-host an open-source runtime can see the options in Paperclip alternatives and in what Hermes Agent is.

Where a hosted squad fits

A hosted squad fits after you have chosen your Claude tools, when the work becomes several recurring roles. Squad (formerly MissionControlHQ) gives you AI teammates that connect to your tools, understand your context, and do work on a schedule. A squad is that group of teammates. Most agent tools give you one.

Squad homepage with the headline AI teammates to do your work and a dashboard preview
The Squad homepage on October 30, 2026.

These facts come from the Squad homepage and docs, checked on November 4, 2026:

Squad costs $99 a month. It includes no AI model usage, so you bring your own AI plan and your provider bills you. Squad adds no per-task AI charges. The docs list 13 providers. Squad recommends Claude Code on a Claude Max or Ultra plan, with Claude Opus 5.5 as the model.

Claude connects in two ways, and they bill differently. The Claude Code option, which the setup page marks Recommended, takes a setup token and, per the card, uses plan limits only and never bills extra usage. The older Claude option is marked Deprecated. It also takes a setup token, and Anthropic bills it as Extra Usage at API prices. Its setup page points you to Claude Code instead.

With Claude Code connected on Max at $100 a month, the arithmetic is $99 plus $100, or $199 a month in two separate bills. On Ultra at $200 a month, it is $299 a month. Claude Code accounts take turns: Squad uses one at a time and moves to the next when the active one reaches its usage limit.

Squad also keeps the human side visible. Everything the squad does is recorded in a run log. You can invite people on paid seats at $9 a month each, and each person gets a role. Teammates can draft for your approval, and you set access levels for anything that must not happen without you.

Squad has a 7-day money-back guarantee on the first payment. It has no free trial. For a comparison of the tools in this space, see why Squad.

When you do not need Squad

You do not need Squad when one person does the work and a single Claude tool covers it. Several cases fit that description.

Squad earns its $99 when several specialist roles need to run continuously, with their own access, and you want one place to see all of it. Squad does not replace Claude Code. Founders keep a coding agent for coding.

Frequently asked questions

Basics

What are Claude agents? Claude agents is a name for several separate Anthropic products. Claude Cowork does non-coding work from a goal you describe. Claude Code is a coding agent. Subagents and agent teams are ways to run several Claude Code agents. Claude Managed Agents is a hosted agent service on the Claude API. The Claude Agent SDK is a library for building your own agents.

What is the difference between Claude Cowork and Claude Code? Claude Code is built for software work in a repository, and it runs in a terminal, an IDE, a desktop app, or the web. Cowork brings the same agentic design to non-coding work such as documents, files, and research, with no terminal. Both are included in paid Claude plans.

Are Claude agent teams ready for production? Not yet. Anthropic labels agent teams experimental and disabled by default, and lists known limits around session resumption, task coordination, and shutdown. Each session can have one team, and a team cannot be shared across sessions.

What is the difference between subagents and agent teams? A subagent works inside one Claude Code session and reports its result back to the main agent. In an agent team, each teammate is a separate Claude Code instance with its own context window, and teammates message each other and share a task list. Teams use more tokens.

Managed Agents and the SDK

What are Claude Managed Agents? Claude Managed Agents is a hosted agent service on the Claude API. Anthropic runs the agent loop, tool execution, and a cloud sandbox, and you create agents, environments, and sessions through the API. It is in beta and needs the managed-agents-2026-04-01 beta header.

How much do Claude Managed Agents cost? Managed Agents bills tokens at the standard model rates plus $0.08 per session-hour while a session is running. Idle time does not count toward runtime. Web search inside a session costs $10 per 1,000 searches.

What is the Claude Agent SDK? The Claude Agent SDK is a Python and TypeScript library that gives you Claude Code's tools, agent loop, and context management inside your own application. You run it in a process you operate and authenticate with an API key.

Plans and limits

Which Claude plan do I need for Claude Code and Cowork? Pro at $20 a month includes both Claude Code and Cowork. Max at $100 a month gives 5x the usage of Pro. Ultra at $200 a month gives 10x the weekly usage of Pro and no 5-hour limit on verified accounts. Team seats and Enterprise include both. The Free plan includes neither.

Do Claude Code and Cowork share usage limits? Usage across claude.ai, Claude Code, and Claude Desktop counts toward the same limit, with a five-hour session window and weekly limits. For Team and Enterprise seats, Anthropic says the allowance is shared with chat and Cowork. Anthropic's usage page does not say how Cowork counts on Pro, Max, and Ultra.

Business use and Squad

Can Claude agents run a business with several specialist roles? Each Claude product covers part of it. Cowork sessions are personal and cannot be shared. Claude Code is built for coding. Agent teams last for one session. Managed Agents and the Agent SDK give you the agent engine, and you build the task board, schedules, approvals, and dashboard yourself.

What is Squad, and how does it relate to Claude agents? Squad (formerly MissionControlHQ) is a hosted service that runs a group of AI teammates, each with its own job, schedule, and permissions, for $99 a month. It uses the AI plan you bring, and it recommends Claude Code on a Claude Max or Ultra plan. Its Claude Code option runs on your Claude plan limits through a setup token.

Do I need Squad if I already use Claude Code or Cowork? No, not if one person does the work in a repository or in personal sessions. Claude Code and Cowork cover those jobs well. Squad fits when several specialist roles need to run on schedules, with their own access and a shared view of the work.

Sources

Last updated: November 4, 2026. Pricing and features verified as of November 4, 2026.