During Apollo, the room full of consoles weighed in on almost everything. One thing it never decided was who flew. Crew assignment belonged to Deke Slayton, the Director of Flight Crew Operations, an executive who answered for that call by name. The room advised. One accountable human decided.
That's the closest picture I have of what running a Digital Crew in Claude Cowork actually feels like.
On Saturday, September 26, between 5:06 and about 8:30 in the morning, the seats on my crew did roughly 15 hours of work, by their own estimates. I did very little of it. My job was to decide. The number I trust most from that morning is a different one: zero, the count of emails the crew sent without me.
If you came here looking for Claude Cowork use cases, this is a real list, one seat at a time, with the mistakes left in. My seats use mission control call signs, so I'll translate each one the first time it shows up.
How the morning started without me
Cowork has scheduled tasks: a job you set up once that runs on its own cadence, as its own session, even while your computer is asleep. Three of mine fired before I sat down. BOOSTER, the seat that runs my sales pipeline, started its daily pass at 5:06. The Content Engine, the scheduled job where three seats work the blog together, started at 5:46. EECOM, the seat that watches the health of my website and channels, started its weekly read at 6:05.
A sales pass that cleared the admin first
Before any selling happened, BOOSTER did the part of sales nobody puts in a job posting. It matched 40 emails to the right contacts, logged 35 touches in my CRM and sorted 52 overdue follow-ups. It held 23 of those back on purpose, because the organizers on the other end were within four weeks of running their own events, and nobody wants a pitch that week. That's Admin Drag, the busywork that buries a seller's week, cleared before six.
Then it got to new outreach, which for me mostly means the people who program insurance conferences and association meetings. Twelve prospects had cleared every automated check BOOSTER runs. It sent none of them.
The reason was a question nobody had answered yet. For eight of those events, did the organizer want a keynote or a breakout session? A pitch for the wrong slot is worse than no pitch, so the crew held all twelve and asked me. Zero cold emails went out that morning, and that was the right number.
It also staged eight follow-ups as drafts, each one sitting in its original email thread and each carrying something new: a link to a byline of mine that had just been published. I read them, and I sent them.
A rule that was wrong, fixed where it lived
Not everything went right. The sales pass proposed opening a deal for someone I'd never sent a quote to. My rule is simple: no deal until a quote goes out. When I said so, the seat traced the mistake back to its own written procedure, which had the rule wrong, and the procedure was rewritten on the spot. My reaction at the time was "No wonder you kept suggesting deals."
That's the moment I'd show a skeptical agency owner. The crew went past the one deal in front of it, found where the bad rule lived and changed it there, and the call on what the rule should say was mine.
Blog posts drafted, checked and held for a human
Inside the Content Engine, the work passes down a short line. GUIDANCE, which plans what gets written and when, picks the post from the plan. PAO drafts it into a review doc; that's the seat that writes everything I publish. OPTICS, the seat that owns every visual, designs the featured image, and then everything waits. Nothing publishes until I approve it by hand. Two posts were approved that morning.
The fastest of them started as a question. Advice I've given clients for years became the Walkdown, my four-move method for redesigning a workflow, and a finished, approved post in about an hour and a half. The crew was more cautious about it than I was, and I overrode it to get there. That's allowed. It's my call.
The mistakes are worth as much as the wins here, because of how they surfaced. One featured image had a headline running off its edge, and it was caught before the image was used. One draft reused a story I'd told two days earlier, and the crew flagged it for me instead of quietly swapping it out. And a gap in one procedure meant I placed a diagram in a post by hand.
A site report that graded itself a fail
EECOM's weekly read covers search rankings, traffic and the list of keywords I'm writing toward. That week it reviewed 122 keywords and retired 11. It also scored a prediction it had made the week before and marked it a miss. I've worked with plenty of people who wouldn't do that, and I've been one of them.
It recommended five new keywords. I asked one question about who was actually searching for them, which led to an intent check, and three of the five turned out to be pulling healthcare practices rather than insurance agents. The recommendation went from five to two.
What I was doing the whole time
Mostly waiting. The conversations ran side by side, all inside Claude. While one seat worked, I moved to the next one, read my morning newsletters, or played a game on my phone. I didn't hop between five apps to do any of it, which takes care of most of the Toggle Tax, the cost of switching between systems.
Two honest limits. The 15 hours is each seat's own estimate of what the work would have taken me solo, a range of 14.5 to 17, and I haven't timed myself doing it by hand. And my attention was split across several chats at once. It didn't feel like juggling, because most of each chat was waiting for me, but it isn't the same as one person doing one thing.
The use case that matters most
Every seat on that list did real work. What makes the crew trustworthy is the short list of things it never does on its own. It doesn't send email, it doesn't publish, and it doesn't decide which events get a pitch. Each seat has a named human owner, which is me, and a written list of the decisions it never makes. That's the difference between a Digital Crew and a stack of subscriptions.
If you run an agency and you're wondering whether to let AI anywhere near client email, start there rather than with the tool. Write down the one thing your first seat will never do without you. Then hire it, and put your name on it.
If you'd like help working out which seat your team should hire first, book an AI strategy call.
Frequently asked questions
Does Claude Cowork send emails or publish on its own? It works with whatever connectors and tools you give it, so where it stops is a decision you make up front. Mine stage drafts and wait, and I do the sending.
Does my computer have to be on for scheduled tasks to run? No. Anthropic's help page says scheduled tasks keep to their cadence even when your computer is asleep or the desktop app is closed, and each one runs as its own Cowork session.
Which Claude plans include scheduled tasks? All the paid ones: Pro, Max, Team and Enterprise, per the same help page.
