Claude Cowork: The First Rung of Agentic Work for Marketing Teams
Anthropic pushed Claude Cowork to web and mobile in beta. Here is what the release means for marketing teams, and why it is the first rung of agentic work.
Key Takeaways
- Claude Cowork is Anthropic's agentic workspace where Claude owns whole tasks in the background rather than trading turns in a chat window, and the new beta extends it from desktop to web and mobile for Max users, with tasks persisting across devices.
- Anthropic reports that over 90 percent of Cowork usage is non-coding knowledge work, led by business operations and content creation, which makes this the most credible on-ramp yet for a non-technical marketing team to start delegating real work to an agent.
- For a CMO, the practical read is that Cowork is the first rung. Once your team gets comfortable handing whole tasks to Claude on their phone, the jump to coding agents like Claude Code becomes a training question, not a cultural one. (What that training actually involves: it.s a context problem, not a prompt problem.)
- Start with the tasks your team already writes long Loom videos to explain: weekly reporting pulls, competitor sweeps, first-draft briefs. Those are the shape of work Cowork is built to own.
- The wrong time to adopt this is when your team has no shared way to describe how work gets done. Agents amplify whatever process you point them at, including the broken ones.
Anthropic just pushed Claude Cowork to web and mobile in beta for Max users, and the framing in the release matters more than the ship date. Cowork is a workspace where Claude owns a task, works in the background, and reports back when it is done, which is a very different animal from the chat window most marketers picture when they hear the word Claude. The desktop version has been out for a while, mostly in the hands of the kind of person who was going to try it first anyway. The web and mobile release is the part that changes who can actually use this at work.
The number that jumped out at me in Anthropic's post is that over 90 percent of Cowork usage is non-coding knowledge work, with business operations and content creation leading the mix. That is a marketing and ops story dressed up as an engineering release, and it is the piece most of the coverage is going to miss, because the headline reads like another agent launch from another AI lab.
My read is that this is the easiest way a non-coder on your team is going to start delegating real work to an agent this year. And the teams that get comfortable delegating in Cowork are one short step from coding agents. That is the arc worth planning against.
What Claude Cowork actually is, in plain terms
The mental model most marketing leaders have for AI right now is still the chat window. You type, it answers, you copy the answer somewhere useful. That loop is fine for asking questions, and it is a terrible fit for actual work, because most work is not one question. Most work is a sequence: pull the numbers, compare them to last month, draft the summary, flag the two things that look off, put it in the shared doc. Chat forces you to conduct that sequence one turn at a time, which is why everyone who has tried to run a real marketing task through a chat window has ended up feeling like a very expensive keyboard.
Cowork inverts that. You describe the task once, Claude spins up its own environment to work in, and it goes. It can browse, read files you give it, run code if it needs to, and keep working while you go do something else. When it is done, or stuck, it comes back with the output and a record of what it did. Anthropic's product page describes it as Claude owning the task end to end, and that is the right framing.
Picture the way this actually plays out. A year ago, a Monday morning competitive sweep meant a coordinator opening ten tabs, pulling notes into a doc, and blocking out two hours before the leadership sync at eleven. In the version Anthropic is now shipping, you kick off the task on your laptop at 8:45, walk into a standup at nine, glance at your phone during the coffee break, and by the time you sit down for the sync the workspace has the three moves your competitors made last week and a note flagging the one that maps to your positioning. You did not sit at a chat window and shepherd it through. You wrote the brief once and went to your meeting.
That is the difference between an agent being a tool you sit down at and an agent being a coworker you check in with. Tasks persisting across devices sounds like a small feature note in a release. In practice it is the change that lets a marketing team actually use this without rearranging their day around it.
Why Cowork finally makes the non-coder onboarding believable
Every agent story for the last eighteen months has had the same problem for marketing leaders: the demos are all code. Watch Claude build an app. Watch Claude refactor a repo. That is genuinely amazing if you are an engineer, and it is a wall if you are a CMO trying to figure out where your team plugs in. You end up nodding at the demo and then going back to prompting in a chat window on Monday.
Anthropic saying, in their own release, that over 90 percent of what people are actually doing in Cowork is non-coding work is the piece that reframes this. Business operations. Content creation. Research. The stuff a marketing team does all day. Read that as Anthropic noticing where the product is already landing, not as a pivot away from developers.
The kinds of tasks that fit Cowork well, from what practitioners keep describing in the CMO community, look something like this. A weekly competitive sweep where you want the top three moves from a list of competitors, summarized against your positioning. A first-draft brief from a transcript plus three reference docs. A pull of last month's numbers from a shared sheet, compared against goal, with the two anomalies flagged. A content audit against a taxonomy you already wrote down. None of those need code. All of them are the kind of thing a smart coordinator does in an afternoon, and all of them are the kind of thing a marketing leader is already writing Looms to explain.
The onboarding pitch to a non-coder gets simpler here. You are still writing a brief, which you already do, and then handing it to something that will actually go do the work while you are on a call. That is a shift in ergonomics rather than a shift in skill, which is why this release changes the conversation for a marketing team in a way that the last three agent announcements did not.
The real story: Cowork is the first rung
Here is the part that matters for a CMO thinking two years out. Cowork is not the destination. It is the first rung of a ladder that ends with your team commissioning software to solve marketing problems.
The pattern showing up across the CMO community is that once someone gets used to describing a task well enough for an agent to run it, the next question they ask is inevitable. If Claude can do the weekly report, can it also build me a tiny dashboard that runs the weekly report on its own? That question is the door to coding agents. It is the same skill, applied to a slightly different surface. You are still writing a brief. The brief just happens to produce a small application instead of a document.
We have written before on Inference about marketers moving from designing rectangles to chatting components into existence in Claude Code, and about why marketers should be in GitHub even if they never write a line themselves. Both of those pieces assume a reader who has already gotten comfortable delegating a whole task to an agent. Cowork on web and mobile is the thing that gets a normal marketing team over that first threshold. Once they are over it, the jump to Claude Code stops feeling like a career change and starts feeling like a next feature.
The teams that are going to look untouchable in two years are not the teams whose CMO learned to code. They are the teams whose CMO taught the whole marketing org to delegate to agents, starting with the tasks that already exist, on the surfaces they already use.
How I would actually start on Monday
If I were running this in-house today, I would not roll it out as a program. Programs are how good tools die in marketing departments. I would pick three people on the team who already write good briefs and get them on the Max plan with Cowork access this week.
Then I would pick three recurring tasks that everyone already agrees are painful. A weekly report. A competitor sweep. A first-draft brief from a call transcript. For each one, I would have the person write down the way they currently do it, step by step, as if they were training a new hire. That document becomes the prompt. Anthropic has a short intro course on Cowork on their Skilljar site and a getting-started article in their help center that are both worth reading before you write the first one, because the shape of a good Cowork task is a little different from the shape of a good chat prompt.
After two weeks, get those three people in a room and have them show each other what worked and what did not. That conversation is the whole point. It is where the team starts to build a shared instinct for what to delegate, and that instinct is what carries them to the next rung. If you want a more structured version of that rollout, we wrote up how we sequence this kind of adoption inside marketing teams in our agentic marketing blueprint, which is the same approach we use with our own clients.
When this is the wrong move
Cowork is not the right first step for every team, and it is worth naming that clearly. If your team has no shared vocabulary for how work gets done, an agent will not fix that. It will just produce output faster against a broken process, which is worse than the slow version, because now you cannot see the seams.
The other case where I would slow down is when the tasks you want to delegate touch systems of record you do not fully trust yet. If your CRM data is a mess, do not point an agent at it and hope. Fix the source first, then delegate against it. Agents amplify whatever they are pointed at, including the parts you were quietly working around.
And if your team is not on a Max plan and cannot get on one, this specific release does not change your week. The broader thesis still holds, and the desktop version and team and enterprise variants are worth looking at, but the web and mobile beta is a Max feature for now.
What we are doing next
The move I would make, if I were the CMO, is to migrate the tasks that used to live in internal Looms and SOPs into Cowork tasks one at a time, and watch which ones stick. The ones that stick get promoted into skills the team shares, the same way you would share a prompt library or a design system today. The ones that do not stick are usually a signal that the underlying process was not as clean as you thought, which is useful information on its own.
The web and mobile release is not the story on its own. The story is that the first rung of agentic work just got a lot easier to step onto, on the surfaces your team already lives on. If you do one thing this week, pick the task your team most often writes a Loom to explain, and hand it to Cowork instead.
Frequently Asked Questions
Is Claude Cowork available to everyone right now?
The web and mobile version is in beta for Max plan users as of the latest release, per Anthropic's announcement. Desktop access and team and enterprise variants have their own availability, documented in Anthropic's help center. If you are on a lower-tier plan, you can still use Claude in the normal chat interface, but the Cowork workspace with background tasks is gated to Max for now.
How is Cowork different from just using Claude in a chat window?
The chat window is turn by turn. You ask, it answers, you ask again. Cowork gives Claude its own environment to work in, so it can take a whole task, work on it in the background across multiple steps, and come back when it is done. Tasks persist across devices, which means you can start something on your laptop and check on it from your phone.
Do I need to know how to code to get value from Cowork?
No. Anthropic says over 90 percent of Cowork usage is non-coding knowledge work, with business operations and content creation leading the mix. If you can write a clear brief for a coordinator, you can write a clear task for Cowork. That is the whole skill on the marketing side.
What kinds of marketing tasks fit Cowork best right now?
Recurring, well-defined tasks with clear inputs and outputs. Weekly reporting pulls. Competitor sweeps against a known list. First-draft briefs from transcripts and reference docs. Content audits against a taxonomy. The tasks you already have a Loom or an SOP for are the right first candidates, because the SOP is basically the prompt.
Where does this fit against Claude Code and other agent tools?
Cowork is the on-ramp. It handles knowledge work in a workspace a non-coder can use. Claude Code sits further up the ladder, where the agent is producing software instead of documents. My read is that teams that get comfortable delegating in Cowork will find the jump to Claude Code much smaller than they expect, because the underlying skill, writing a good task, is the same.
