A Director of Brand at a Series-C SaaS told me last week that she tried Claude Cowork on a Monday.
She pointed it at the folder where the team keeps the messaging docs and last quarter’s campaign assets and asked it to find every place they’d referred to the enterprise tier as “enterprise plan” versus “enterprise tier” versus “Pro Max.” Three minutes later it handed her back a spreadsheet — 47 hits, sorted by document age, recommended canonical phrasing in column D. The audit had been on her to-do list for six months. She’d been planning to spend a Saturday on it.
She said she closed the laptop, opened it again, and just sat there for a second.
If you’ve used Claude in chat mode — the paste-a-paragraph, ask-for-a-rewrite, riff-with-it pattern most marketers have used by now — Cowork is a different product than the one you’ve been using. That’s what the Director of Brand’s Monday surfaced for me. Most marketing teams I’ve talked to this month haven’t caught it yet.
Chat and Cowork are different products
A note for any marketer whose team hasn’t tried Cowork yet: Claude Cowork is Anthropic’s autonomous-agent product. You point it at a folder, a tool, or a goal; it works on your machine for a few minutes or longer without you having to steer each step; it hands you back a finished deliverable. Per Anthropic’s own product page, it’s positioned for “non-technical teams” — and the page explicitly cites Anthropic’s own Marketing and Data teams as early adopters who “started bypassing Claude’s chat interface for Claude Code, drawn to its ability to handle complex, multi-step work.”
That’s the part most marketers don’t internalize from reading the page. Cowork is not a better chat interface. It’s a different interaction pattern. Chat is a conversation partner — turn-by-turn, sentence-by-sentence, fast iteration. Cowork is a handoff — you give it a goal and the source material, you let it work, you read the deliverable.
Most marketers’ first Cowork run is a chat-shaped prompt that produces a chat-shaped result. They paste a paragraph, ask for “make this better,” watch it spin for two minutes, get back something fine, and conclude Cowork is overrated. The disappointment is the chat pattern’s fault, not Cowork’s.
A head of demand gen I spoke with this month told me her first Cowork run was a disaster and that it was her fault — she’d treated it like ChatGPT. Then a colleague showed her what a Cowork run looks like when you let it actually be a handoff. The colleague had pointed Cowork at the folder of last quarter’s webinar transcripts and asked it which talking points actually moved registration numbers when used in the follow-up email. It came back with a four-page memo, citation by citation, pulling lines from the transcripts. “That was the moment I understood Cowork is a different thing,” she said.
So the question for a marketing team isn’t should we use Cowork. The question is which marketing jobs does the handoff pattern actually earn its place on. I’ve been keeping a running list. Here’s where it’s landed so far.
Four jobs Cowork is good at
Competitive-research synthesis across many sources. A marketing team’s competitive-research file is usually 20 to 60 documents — competitor websites pulled into PDFs, analyst notes, sales-call summaries, the deck the head of product brought back from the conference. Synthesizing that into a positioning memo is six hours of reading you’ve been pushing off all quarter. Cowork pointed at the folder with the question “what’s the single argument our top three competitors are making that we don’t have a public answer to” is exactly the shape of work the handoff pattern earns. You’re not riffing with it. You’re handing it the reading.
Campaign-asset assembly from source files. Six campaign emails, three landing pages, two social variants, a webinar registration page — all of them carrying the same offer, all of them needing to land slightly differently for slightly different audience segments. The work is mostly stitching: pulling the right product claim from the master messaging doc, pulling the right proof point from the customer-story library, pulling the right call-to-action language from the campaign brief, dropping each into the right asset shape. Cowork pointed at the campaign folder and the messaging library, asked to produce first drafts of all six assets against the brief, is a job that used to eat a marketer’s Friday. The drafts come back at the quality bar of “needs review” — not “ready to ship.” That’s correct. The marketer’s job becomes the review, not the assembly.
Performance-data-to-narrative drafts from your analytics exports. Monthly review. The CMO needs a one-page narrative on why last month’s campaigns performed the way they did. The data is in your analytics tool, your CRM, your email platform, your spreadsheet of UTM-tagged sources. Exporting all of it into one folder and asking Cowork “draft the narrative — what worked, what didn’t, what’s the one thing I should change next month” is the kind of synthesis that used to take a head of growth four hours of pivot-tabling. Cowork’s first draft will be wrong in at least one place; the marketer who knows the underlying business is the one who catches it and corrects it. That’s fine. The point was to skip the pivot-tabling, not to skip the judgment.
Brand-voice audits across past content. The Director of Brand at the open is in this category. The audit was “find every place we’ve used inconsistent product naming, and tell me which canonical version we should standardize on.” It was 47 hits across two years of content. Three minutes. The marketer’s job is to decide which canonical phrasing wins — Cowork’s job was the labor of finding all 47.
There’s a pattern across the four. None of them are creative work. None of them require the marketer to riff. All of them are work that has a clear deliverable, a definable source set, and a known shape — and the bottleneck is that one human would have to read or stitch a lot of material to produce the deliverable. The handoff pattern earns its place when the labor-to-judgment ratio is high.
Two jobs where chat is still the right tool
I want to name these explicitly because if I don’t, the rest of the piece reads as a Cowork pitch and I lose your credibility.
Live brainstorming with a colleague. A CMO at a Pavilion breakfast in May said the thing I’d been wanting to put words on: “Cowork can’t replace the brainstorm I have with my head of content on a Tuesday. Two of us, a whiteboard, an hour. That’s not a Cowork job. Cowork is for the part where afterwards someone has to go through 40 customer-interview transcripts and find the three insights the brainstorm assumed were true.” Chat with a colleague is still a different category from chat with an agent — and chat with an agent (Claude in chat mode, riffing fast) is still a different category from a Cowork handoff. The marketer who tries to replace the Tuesday brainstorm with a Cowork run produces a worse Tuesday and a Cowork output nobody wanted.
Single-shot writing of a single paragraph. “Rewrite this email subject line.” “Tighten this opening paragraph.” “Give me three variants of this CTA.” That’s a chat job. The setup time of pointing Cowork at source material, defining a deliverable, waiting for the run, costs more than the rewrite would have. Chat is fast, narrow, conversational. Use chat for fast, narrow, conversational work. Use Cowork for the stitched-together synthesis that would otherwise eat a half-day.
The line isn’t about quality. Chat and Cowork are both Claude. The line is about interaction shape. If what you need is iteration, use chat. If what you need is a deliverable from a defined source set, use Cowork.
The upstream variable
Here’s the part most pieces about Cowork don’t say.
Cowork’s quality on any given marketing job is bounded by what it can reach on the machine where it runs. The Director of Brand’s three-minute audit worked because the messaging docs were already on her disk in a folder she could point Cowork at. If the brand voice rules had lived only in her head, the audit would have produced a list of occurrences with no recommended canonical phrasing — useful but half the job. If the campaign-asset library lived in a different teammate’s Drive that her machine couldn’t read, Cowork would have produced first drafts that didn’t match the current campaign’s voice. The output is downstream of what’s reachable.
This is also why marketing teams that are doing this well are not the ones that bought Cowork licenses. They’re the ones whose function heads — the Director of Brand, the Director of Demand Gen, the Director of Content, the Director of MarOps — have been quietly cleaning up what lives on the team’s shared disk for the last six months. The brand voice rules in one place. The current quarter’s positioning in one place. The customer feedback themes summarized in one place. The campaign performance data exported and named consistently in one place. None of it was for Cowork specifically. It was good hygiene. But it turns out good hygiene is exactly what makes the difference between a Cowork run that produces generic and one that produces something that sounds like your team.
If you’re the CMO, this is the thing to notice: you don’t need to centralize what Cowork can reach in your seat. Your function heads are the ones closest to their slice of the substrate. The Director of Brand owns what the brand voice rules look like; the Director of Demand Gen owns what the campaign performance data looks like; the Director of Content owns what the past-content library looks like; the Director of MarOps owns the disk hygiene that makes any of it reachable. They are the curator-owners of their function’s slice. You’re not delegating this; you’re giving it to the people who are already closest to it.
Pomegranate, the piece I’m building, is one way to make that practical — it carries your team’s voice and current direction across every AI tool your team uses, including the Cowork runs that read from local files. There are other ways to set this up; the point isn’t the tool. The point is that until the function-head slices of the substrate are curated and reachable, every Cowork run anyone on your team does is bounded by whatever happened to be on the local disk that day.
The ongoing cost is small but real: about five to ten minutes a day from each function head, on average, curating their slice as new decisions land and new themes emerge. Hide that cost and the substrate decays; name it and the next Cowork run reads from material that’s actually current.
You’ll feel the difference first in the deliverables. The feedback you’d give, already in the draft. The Monday Cowork run starts coming back at the quality of someone who’s read everything your team has decided this quarter — because it actually has.