Abel

I'll be offline from tomorrow until Wednesday. It either ships today or it doesn't.

The system: unique Stripe coupons per user, triggered on signup, tied to a specific credit amount. Not complicated in theory. The part that takes time is the edge cases - what happens if someone's already signed up, if the coupon's been used, what the flow looks like.

I'm building a system today that needs to be done before I go offline tomorrow.

The honest argument for operators learning Claude + MCPs isn't that you'll replace analysts. It's that you'll ask ten questions you would've swallowed before.

What I didn't expect. Once the report ran, I wanted to see the next-biggest cluster. Then the one after that. I'd never asked any of those questions before because the friction of asking an engineer was higher than the curiosity of wondering.

I found a user with 17 accounts last week. I'm not a data analyst. I built the detection in an afternoon.

The result: The morning ritual is gone. Refund request volume went down, too. I think that's because the apologies sound human, and people who feel heard don't escalate.

The part that took the longest was the apology email. I didn't want it to sound like a template. I had Claude read 20 of my teammate's past apology emails, then write a new one that sounded like her. She edits one sentence before sending.

The most valuable internal tools aren't the ones that replace work. They're the ones that unstick the small repeating steps you never called work in the first place.

If you're a generalist on a small team and you outsource every data question to whoever on eng is closest, stop. Spend an afternoon connecting Claude to the tools you already pay for.

What I thought would happen: I'd spend an hour fighting with the tool. What actually happened: The hard part wasn't the data. It was finding the users' emails because they were on separate org accounts - which is its own problem we now have to fix.

I could say 'tell me when they're going to run out of credits and which users on the team are burning through the most,' and Claude would write the query, run it, read the result, and turn it into a paragraph.