how i killed a $1,000-a-week task with claude and a gmail connection
A team I work with was about to pay a virtual assistant a thousand dollars a week to sit in their inbox and draft replies. Not close deals. Not write strategy. Just read incoming email and write first-draft responses so the founders didn’t start every reply from a blank screen.
A thousand dollars a week. Fifty-two thousand a year. To draft email.
I get why. The inbox is real work and nobody wants to do it. But before they signed that check, we tried something else. It took an afternoon to set up and it does the same job for almost nothing.
what we actually built
We connected Claude to the team’s Gmail. When a new email comes in, Claude reads it, pulls the context it needs, and writes a draft reply in the founder’s actual voice. The draft sits in the inbox waiting. A human reads it, fixes anything that’s off, and hits send.
That’s the whole thing. No new platform to learn. No dashboard. The work shows up where the work already lives, which is the inbox.
The part that matters is the voice. A generic AI reply sounds like a generic AI reply, and people can smell it. So we fed Claude a stack of emails the founder had actually written. Now the drafts sound like him. Short where he’s short. Warm where he’s warm. The reader can’t tell a machine started the sentence.
why this is the real version of “AI for your business”
Here is the test I run on any AI idea before I build it: does this save real time or real money, today, without making the work worse?
This one does. Same output, a fraction of the cost, and the founder still reads every reply before it goes out, so quality control never leaves the building. That last part is the difference between automation that helps and automation that quietly embarrasses you.
Compare that to what you see online. Somebody built an app in half a day and it’s already making a million dollars. That is not a strategy. That is a person trying to go viral. The real wins are boring. They look like an inbox that drafts itself overnight and a VA invoice that disappears.
what it does not do
It does not send anything on its own. I would not let it, and you shouldn’t either. The value is in the first draft, not the send button. A person stays in the loop because your name is on the email and trust is the only thing you actually sell.
It also will not fix a process that is broken underneath. If your team can’t answer a question because nobody knows the answer, a faster draft doesn’t help. AI makes a good system faster. It makes a broken system fail faster.
the math, plainly
- Old way: roughly $52,000 a year for a person to draft replies.
- New way: a setup that costs an afternoon and a few dollars a month in usage, with a human reviewing every draft.
That is the kind of swing that is sitting in most small businesses right now, unclaimed, while everyone argues about whether AI is going to take their job. It is not going to take your job. It is going to take the thousand-dollar-a-week task you hated anyway.
want to find yours
Most teams have at least one of these. A repetitive, low-judgment task someone is paying full price for. If you want, I’ll help you find it and tell you honestly whether AI should touch it. Sometimes the answer is no. That’s worth knowing too.