Sanaf and Claude Cowork: a shared desk against a shared employee

Claude Cowork is Anthropic's agent for everyday knowledge work. You hand Claude a task that takes more than one answer, such as a report built from a folder of files, a deck, or a weekly summary, and it plans the steps, opens the files and tools it needs, and works through them while showing you each step. On 16 September 2026 Anthropic stopped selling it as a separate mode. In the company's words, "Claude Cowork and chat are merging into one Claude", so the same window now decides whether you want a quick answer or a longer piece of work.

That merge changes what people mean when they search for Claude Cowork, and it sharpens the question this post is about. Cowork is very good at doing work for the person signed in to it. A lot of businesses want something slightly different: a colleague the whole team can hand work to, one that holds a job and keeps doing it whether or not anyone opens an app that day. That is what an AI employee like Sanaf is for. Below is what Cowork is today, what it costs, where it is the better choice, and where it isn't.

Everything about Anthropic's product comes from its own live pages, read on 30 September 2026: the Cowork product page, the pricing page, the merge announcement and Anthropic's Cowork help article. We have not run a trial of our own for this post, so nothing here claims a hands-on result.

What is Claude Cowork?

Cowork started life as a research preview in January 2026. Simon Willison wrote up his first impressions on launch day, 12 January, when it lived inside the Claude desktop app for Mac and was open only to subscribers on Anthropic's $100 and $200 individual plans. The idea was to take the agent behind Claude Code, Anthropic's coding tool, and point it at ordinary office work. You gave it a folder, and it could read, edit and create the files inside. Pro subscribers got access a few days later.

Nine months on, the product page describes something much bigger. Claude "runs tasks in your folders and tools", splits big projects into chunks that run side by side, and keeps going after you close the laptop. It can open its own browser in a side panel while you keep using yours, and it can import your cookies so it stays signed in to the sites you use. It connects to other apps through connectors (Slack, Google Drive, Microsoft 365 and Amplitude are among those listed) and plugins. And it shows "each step: the files it opens, tools it uses, and choices it makes."

The help article adds a detail that matters for anyone planning around it. Cowork now "runs your tasks in the cloud (in beta)", on Anthropic's servers in an isolated environment. Sessions keep running when the desktop app is closed or the computer is asleep, but "if your task uses local files, your browser, or your computer, keep the desktop app open so Claude can reach them."

How to use Claude Cowork after the merge

There is no separate tab to find any more. Since 16 September you ask in the normal Claude window and it decides how much machinery the job needs. The announcement gives the recurring version plainly: "Schedule it for every Monday, and Claude starts on the report without being asked." Two newer pieces, Claude Docs and Claude Slides, let it write documents and presentations inside the conversation, and TechCrunch's coverage of the merge notes you can download the slides as a PDF or PowerPoint.

The rollout is staged. Anthropic says the merged Claude is rolling out to its individual paid plans (Pro and the higher tier above it) "over the next few weeks, with more plans to follow", with Team and Free after that and at least 30 days' notice for Enterprise admins. If you are on a Team plan and still see a Cowork tab, that is why.

Claude Cowork scheduled tasks

Scheduling is the feature that moves Cowork from assistant toward coworker. The product page says you can "schedule a task for any cadence, and it runs unattended", and the help article says scheduled tasks run in the cloud, so they "don't need your computer to be awake or the desktop app open." For a person who wants the same Monday report every week, that is exactly the right feature.

Claude Cowork vs Claude Code

These get confused because they share an engine. Claude Code is Anthropic's agent for software work: it lives in a terminal or an editor, reads a codebase, runs commands and makes changes to code. Cowork was built from the same agent for people who don't write code, so it works on documents, spreadsheets, slides, browser tasks and connected apps instead of a repository.

After the merge, Claude Code is still its own product and still listed separately on the pricing page. Cowork is the part that disappeared into ordinary Claude. If you are a developer, you want Code. If you want a report written from twenty files, you want what used to be called Cowork, which you now reach by just asking Claude.

Claude Cowork pricing: is Claude Cowork free?

No. The Cowork product page says it "works with a paid plan", and the help article lists every paid plan: Pro, the higher individual tier, Team and Enterprise. Here is what the pricing page showed on 30 September 2026:

Two things on that page shape the real cost. First, "every plan has usage limits that reset on a rolling five-hour session window, and paid plans add weekly limits on top." Second, the help article warns that multi-step work, the kind Cowork exists for, "use[s] more of your usage than a quick question." So whether Pro is enough, or you need the $100 tier, depends on how much agent work you run. Someone who hands Claude a long task most days will hit Pro's limits sooner than someone who mostly chats.

The Claude Team plan is where a business usually lands, and it is per seat. Every person who wants to hand work to Claude needs a seat of their own.

Claude Cowork use cases where it is the right choice

Cowork is strongest when one person owns the work and wants a very capable pair of hands at their own desk. Some examples that fit its design well:

• Turning a folder of contracts, invoices or research notes into one summary or spreadsheet.
• Building a deck or a written report from material you already have, then revising it with you in the same conversation.
• A weekly report you would otherwise assemble by hand, scheduled for Monday morning.
• Browser chores on sites you are already signed in to, run in Claude's own browser panel while you do something else.

If that describes your problem, and especially if your company has already standardised on Claude, Cowork is hard to argue against. The model is excellent, Anthropic shows its steps, and "by default, Claude asks before taking an action", with an option to let it run and check in only when something needs a closer look.

Where a desk and an employee part ways

The title of this post calls Cowork a shared desk. That is meant kindly. Cowork is a very good desk: your files, your browser, your connected apps, and an agent working through them on your behalf. What it inherits from that design is that it works as you, for you.

For a team, that shows up in four places.

Who can hand it work

A Cowork task belongs to the account that started it. If the office manager schedules the Monday report and then goes on holiday, the report is tied to their account and their connections. There is no shared colleague for the rest of the team to message.

Whose sign-ins it uses

Cowork reaches your apps through the connectors and browser cookies of the person using it. That is convenient for one person and awkward for a business, where you usually want the thing chasing invoices to have its own access, limited to that job, and to keep working when someone leaves.

Where it lives

Cowork lives in Claude's own apps on the web, desktop and mobile. Slack is a connector it can read from and post through. Microsoft Teams isn't something the product page lists as a place it works.

How the cost grows

Seats. A ten-person team on standard Team seats billed monthly is $250 a month before anyone picks a premium seat, and the usage limits still apply per person.

None of these are defects. They follow from Cowork being an agent for a person. They only matter if the work you want done belongs to the team rather than to one person.

Sanaf as a Claude Cowork alternative

Sanaf is one AI employee that a business hires into the Slack or Microsoft Teams it already uses, or talks to in its own chat in the console. You give it jobs from a roster of thirty five, such as answering the support inbox, following up leads before they go cold, chasing overdue invoices or running someone's calendar. Each job comes with a written description of what it does.

The differences from Cowork line up with the list above.

Anyone on the team can hand Sanaf work, in a channel or a direct message, and the job doesn't belong to whoever set it up. A job starts when a person asks, when a schedule fires, or when something arrives in an app Sanaf watches, such as a new lead in the CRM or an invoice going past due.

Sanaf keeps its own sign-ins, each one scoped to the job that needs it, across the apps you already run. Nobody's personal browser session is involved, and the access stays put when people change roles.

It asks before anything it cannot undo. You set rules per kind of action (do it, ask first, or never), and it drafts and asks until a person has said otherwise. It keeps a record of what it did and why, so the manager who wasn't in the thread can still see what happened.

It is priced for the work. You start with $120 of free credit and no card. After that you buy credits when you want them, from $20, and they never expire. Adding people or jobs adds no seat fee, and nothing renews on its own. The pricing page has the details.

And if you like Anthropic's models, you don't have to give them up. Sanaf routes work to named model providers through one gateway, Anthropic among them, and none of them train on your data.

A side by side

Which one should you pick?

Pick Claude, with what used to be Cowork, if the work is yours: your documents, your decks, your research, your recurring report, and you want the strongest general agent available at your own desk. Many people who hire Sanaf keep a Claude or ChatGPT seat for exactly this.

Pick Sanaf if the work belongs to the team and nobody has time to drive it: the inbox that's still full on Friday, the leads that go quiet, the invoices nobody chases. It works in the Slack or Teams you already have, everyone can reach it, and you pay for what it does instead of for how many people might ask.

If you are weighing Anthropic's own Slack teammate as well, our Claude Tag comparison covers that one, and the ChatGPT comparison covers the other assistant most teams already pay for.

Questions people ask about Claude Cowork

Is Claude Cowork still available?

Yes, under a different name. Since 16 September 2026 Anthropic has been folding it into ordinary Claude on the individual paid plans, with other plans to follow, so you ask Claude directly instead of opening a Cowork tab.

Does Claude Cowork work when my computer is off?

Partly. Anthropic's help article says tasks now run in the cloud in beta and scheduled tasks don't need the computer awake, but a task that uses your local files, your browser or your computer needs the desktop app open.

Is there a Claude Cowork alternative for a whole team?

If you want the same kind of agent for each person, a Claude Team plan gives everyone a seat. If you want one colleague the whole team shares, with its own sign-ins and a job it keeps doing, an AI employee like Sanaf is built for that, and it works in Slack and Microsoft Teams.

Does Sanaf use Claude?

Among others. Sanaf sends work to named model providers through one gateway, and Anthropic is one of them. Your data is never used to train any of them.

See every job Sanaf can hold, check what starting free costs, or read the Claude Tag comparison if Anthropic's Slack teammate is also on your list.