On 16 September 2026 Anthropic merged Cowork into Claude. There is no Cowork tab any more. You ask Claude for something and it decides whether that is an answer or a task to work through on its own. The merged version is rolling out to Pro and Max first, with Free and Team “later”.

That makes Cowork the default agent for everyone who pays for Claude, and it is a good one. Hand it a folder and an outcome and you get back spreadsheets with working formulas, decks and reports. Scheduled tasks run in Anthropic’s cloud, and Claude in Chrome reaches the sites you are logged in to.

The reason people look past it is usually not quality. It is one of these:

  • More than one person needs to delegate. Cowork is built around one person’s sessions. Team plans can share a project with view or edit access, but there is no list where a colleague and the agent hold the same task and everyone sees who did what.
  • They want more than one vendor’s models. Every step runs on Anthropic models.
  • The work lives on their own machines. Cowork runs in Anthropic’s sandbox with the folders you grant it, not on your server or in your repo.

This post ranks the alternatives on those three. Disclosure: TODO for AI is ours. It is ranked first for teams and not for the other cases, and its limits are in its own section.

The short answer

If you needPick
A team handing work to the AI on one shared listTODO for AI
The same thing as Cowork, from OpenAIChatGPT Work
An agent inside Outlook, Teams and OfficeCopilot Cowork
One-off research and decks, no setupManus
A fixed process that runs the same way every timeLindy
Everything on your own server, for freeHermes Agent

1. TODO for AI: best if your team needs more than one person delegating

A to-do list for the business where each item is done by an agent. Everyone on the team adds work to the same board, and the card shows who asked for it, which agent ran it, and what came back. Instead of one person’s chat history, the team’s work is on a list anyone can check.

What it does that Cowork does not:

  • One board for people and agents. A colleague can pick up an item the AI started, or the other way round. Filter the board by person. How the team board works.
  • Supported models across vendors. Claude, GPT, Gemini and Grok can work from one credit budget, chosen per agent. One model can review another’s work; consumption varies by model.
  • Your own machines and browser. A small bridge gives the agent a shell on your PC or server, a Chrome extension works in your real logged-in browser, and a cloud VM keeps running when your laptop is closed.
  • One workspace. One team budget. Standard $30/month, Premium $120, minimum two seats. Each adds 30 or 120 monthly credits, plus 50 per team. Cloud VM and browser included. Personal session and weekly limits apply. Pricing, credit rates and limits.
  • Company context, voice and memory. Agents start from shared business notes and documents. A report, a customer reply and a website change can use the same company context, with the writing voice set for each channel.

Where it is weaker: Claude’s document output is better polished, and Cowork needs no setup. Here you install the extension, and the bridge if you want it on a machine. Browser tasks need your Chrome or Firefox open. Permissions are per tool (allow, ask first, block), so it can stop before any stripe command but cannot tell a refund from a balance check. It is also a much smaller company than Anthropic.

Free to start with $2 of usage on Claude Sonnet. Starter is $20 with the supported model catalog. Full comparison with Cowork.

2. ChatGPT Work: best like-for-like replacement

OpenAI’s answer to Cowork, launched on 9 July 2026. It replaced ChatGPT agent, which OpenAI has since retired. Work takes on long, multi-step tasks and hands back finished sheets, slides, docs and web apps. It runs scheduled or triggered tasks, and in the desktop app it can use your local files and apps through computer use.

Choose it if your team already lives in ChatGPT, or you prefer GPT’s output to Claude’s. The trade-offs mirror Cowork’s. It runs OpenAI models only. Its cloud browser keeps its own sessions, so you sign in there instead of using your Chrome. Team sharing means sharing a scheduled task that a colleague then copies for themselves. Usage “varies with the amount of work required” and draws on your plan’s allowance. Plus is $20 and Business is $25 per seat.

3. Copilot Cowork: best if your company runs on Microsoft 365

Microsoft’s agent, generally available since 16 June 2026. It drafts and sends mail, builds the Excel model and the deck, posts in Teams and files things in SharePoint, with the permissions and audit trail your IT admin already has. It runs on Anthropic models, the same family as Claude’s Cowork, with GPT in Microsoft’s Frontier preview.

It needs a Microsoft 365 Copilot license first ($18 per user per month on a yearly promotion that ends on 31 December 2026, $21 after), then usage in Copilot Credits at $0.01 each. It is the best choice inside Microsoft 365 and a weak one for anything outside it. Our Copilot Cowork alternatives post goes deeper.

4. Manus: best for one-off research with no setup

Give it a goal and it browses, researches and hands back a document or a deck. No install, and a Browser Operator extension if it needs your logged-in browser. Its desktop app can work in local folders you approve, and since September 2026 Manus Flex lets you bring your own model key; by default Manus picks the model. Pricing is in credits, from $20 for 4,000, and a heavy task can use a lot of them. Good for the occasional project, less so as the place a team’s work lives. Comparison.

5. Lindy: best for repeatable processes

You build an “AI employee” from a trigger and steps: triage the inbox, update the CRM, prepare the meeting. Then it runs every time. Where Cowork is best at open-ended tasks, Lindy is best at the same task done the same way for the fiftieth time. From $29.99 for 3,000 credits, pooled across the workspace. Comparison.

6. Hermes Agent: best free, self-hosted option

An open-source agent you run on your own machine or VPS and talk to from Telegram, WhatsApp or Slack. Full shell, any model with your own API keys, persistent memory, and skills it writes itself. MIT licensed and free apart from tokens. You operate it: updates, keys, sandboxing, keeping it off the public internet. It suits a technical person, not a team of non-technical colleagues. Comparison.

Open source Cowork clones. If free and open source matters more than the team, look at OpenWork (a desktop app built on OpenCode, any model including Ollama, imports Cowork skills and MCP servers) and Eigent (Apache-2.0, local desktop). Both run on your own files. More in best open source AI agents.

Side by side

Team shares the workModelsRuns whereYour logged-in browserFrom
Claude (Cowork)Projects shared on Team, tasks per personAnthropic onlyAnthropic’s sandboxClaude in Chrome$20
TODO for AIOne board, people and agentsEvery vendorYour PC, your servers, a cloud VMChrome extensionFree, $20
ChatGPT WorkScheduled tasks can be shared as copiesOpenAI onlyOpenAI’s cloud, desktop appCloud browser, own sessions$20
Copilot CoworkPer user, inside Microsoft 365Microsoft picksMicrosoft’s cloudEdge, preview$18 + credits
ManusTasks shared one by oneVendor picks, or BYOK via FlexTheir cloud; local folders via desktop appBrowser Operator extension$20
LindyShared workflows and creditsYou pickTheir cloudComputer use$29.99
Hermes AgentChat threadsAny, your keysThe box you hostSelf-configuredFree + tokens

The same comparison for more tools and six criteria is on the alternatives table.

How to choose

Is it one person delegating, or the business? If it is you and your own tasks, stay with Cowork unless one of the other two reasons applies. If three people need to hand work to the AI and see what happened, you need a shared list, and that is what TODO for AI is built for.

Does the work stay inside one vendor’s world? All-Microsoft means Copilot Cowork. All-ChatGPT means ChatGPT Work. Work spread across a CRM, a shop admin, an ad account and a server needs something that works in your browser and on your machines.

Is it the same task every week? Then build it once in Lindy.

TODO for AI team credit terms were checked on 2026-10-09. Other vendor figures retain their 2026-09-24 check; verify current limits and promotions before purchasing.