ooligo

Claude Cowork

ai-agent-platform ai-agents · ai-assistant · workflow-automation · computer-use
AI-NATIVE MCP
RevOpsLegal OpsRecruiting & TACustomer Success
8.5 /10

What it is

Claude Cowork is Anthropic’s agentic mode for knowledge work that isn’t coding: research, analysis, document building and other multi-step tasks. You give it a goal instead of a prompt, and it works across your files, connected apps and a built-in browser until the output exists. It is no longer a separate tab. On 16 September 2026 Anthropic merged Cowork and chat into one Claude. The product page now reads “Claude Cowork is now just Claude.” The merged app became generally available on Enterprise on 23 October and on Team on 12 November, and is rolling out on Pro, Max and Ultra. The Cowork name survives in the help center and in the plan bullets. Underneath, it is now the part of Claude that does work rather than answers questions.

The second change is where the work runs. Since 6 October 2026, new Cowork tasks on Pro, Max and Ultra run in the cloud, and the “Only on your computer” option is gone for good. Local files, browser control and computer use still need the Claude desktop app for macOS or Windows to be open and connected. Anthropic points local-only work to the Claude Code desktop app.

The nearest entries in this catalog are Lindy and Dust, agent platforms sold to operators. Cowork is not a builder. There is no canvas and no agent to configure first. You describe the outcome and steer while it runs.

Why it shows up in ops stacks

The reason is distribution, not capability. Cowork is included on every paid Claude plan. An ops leader whose company already bought Claude seats for engineering therefore has an agent surface for the finance analyst and the recruiting coordinator, with no new procurement or vendor review. The merge strengthens this: anyone who opens Claude now lands on it.

Three capabilities do the actual work:

  • Scheduled tasks run remotely. On every paid plan, /schedule or the Scheduled sidebar sets an hourly, daily, weekday or weekly job. It runs when your computer is asleep or the desktop app is closed. The exception is a task that needs local files or apps, which runs only on the machine.
  • Plugins bundle skills, connectors and sub-agents into one install, and a plugin added once works in chat, Claude Code and Cowork. Since 24 February 2026, Team and Enterprise owners can publish a private plugin marketplace and mark plugins as installed by default, required or blocked.
  • Connectors before the screen. Claude tries a connector first, then the browser, and touches the screen last. Slack, Microsoft 365, Google Drive, Notion and the rest of the connector directory work on desktop, web and mobile.

Pricing reality

Cowork has no price of its own. You buy a Claude seat, and every paid plan includes it:

  • Pro: $20/mo, or $17/mo billed annually
  • Max: $100/mo at 5× Pro usage, now with $100/mo in API credits
  • Ultra: $200/mo at 10× Pro weekly usage, no 5-hour limit for verified accounts, $200/mo in API credits
  • Team: 2 to 150 seats. Standard seats cost $25, or $20 on annual billing. Premium seats cost $125, or $100 annually, for 5× the usage. Up to $500/mo in API credits is pooled across the team.
  • Enterprise: $20/seat/mo billed annually, 20-seat minimum, plus usage at API rates

The tier list is not the real number. Anthropic publishes multipliers, not token quotas. Limits are metered per plan, on a 5-hour window and a weekly window, and there is no separate Cowork allowance. Anthropic’s own product page warns that Cowork “consumes limits faster than Chat.” Treat the $20 seat as an evaluation seat and budget $100–$200 per heavy user in production. On Enterprise there is no seat ceiling on spend: usage bills at API rates, so set user and org spend limits before rollout.

Best for

Ops leaders in RevOps, Legal Ops, Recruiting and Customer Success at companies already paying for Claude who need to hand a non-engineer a recurring document-and-spreadsheet job that runs unattended. Typical jobs: a weekly pipeline-hygiene report built from CRM exports in Drive, a monthly diff across a folder of vendor contracts, a Monday candidate digest assembled from ATS exports and Slack threads. It wins outright on recurring, file-heavy work with a stable output format, owned by one named person, over data that person is already entitled to see.

Not for

Cowork acts as you: your seat, your grants, your connector permissions. It has no per-agent service identity and no audit subject distinct from the human who owns the session. Anthropic says its OpenTelemetry stream of Cowork events “doesn’t replace audit logging for compliance purposes.” If you need a registry of agents with their own lifecycle, look at Microsoft Agent 365 or Salesforce Agentforce. It is also the wrong tool for high-volume structured pipelines. For “process 5,000 rows the same way every night”, an n8n flow or a direct API call is cheaper and more predictable.

Versus the alternatives

  • ChatGPT Work: OpenAI’s equivalent, with “more than 1,400 plugins”, a Plan mode that drafts a step-by-step plan for you to approve, and scheduled and team tasks. It runs on desktop on every plan, and on web and mobile for Plus, Pro, Business, Enterprise and Edu. Pick it when connector breadth is the binding constraint or the organization has standardized on OpenAI.
  • Microsoft 365 Copilot: the incumbent by seat count in ops orgs, $30/user/mo enterprise or $21 on Business, with Agent Mode building whole documents in Word and Excel. Pick it when the work lives in Office files on SharePoint and governance is a tenant-policy conversation.
  • Manus: the fastest-growing entrant. Manus 2.0 (28 September 2026) added a Cloud Computer, event-triggered Automations and computer use. Pick it for open-ended research-and-build tasks from a one-line brief. Check the ownership history first: Meta’s purchase was unwound by a Chinese regulator, and Tencent-led investors now control the company.

Watch-outs

  • Computer use is Pro, Max and Ultra only. The help center is explicit: “Team and Enterprise plans don’t have access to computer use at this time.” Even there it is a beta, limited to macOS and Windows, and not in the Linux beta. Buyers who evaluate on a personal Pro seat and roll out on Team seats find the demo stops working. Guard: before buying seats, split target workflows into the ones that need screen control and the ones a connector reaches, move everything you can onto connectors, and keep the remainder on individual seats budgeted separately.
  • The cloud default changed the data path. Pro, Max and Ultra tasks now run on Anthropic’s servers whether or not you meant them to. On Team, “Run Cowork in the cloud” is on by default; on Enterprise it is off. In HIPAA-enabled organizations, Cowork is off by default, the BAA covers only the local-mode configuration, and cloud sessions are unavailable. Guard: decide where each workflow’s data may be processed before rollout. On Enterprise, grant cloud sessions only to groups that have cleared that review, and route regulated work to local mode or Claude Code desktop.
  • The sandbox limits network reach, not your data. A session runs in an isolated environment that “can’t reach your home or company network” and is removed at the end. But isolation “doesn’t change what Claude can read or do through the access you’ve granted”, and network egress settings do not apply to web fetch, web search or MCPs. A scheduled task that reads a public page is running untrusted text on a timer with your permissions. Guard: grant the narrowest folder that holds the job’s inputs, never pair a web-reading step with a write-capable connector in one scheduled task, and keep writes on manual approval. On Enterprise, turn on skill scanning and switch off web search for groups that don’t need it.