Sold a $700 app to a coffee shop. I didn't write it, Claude did.

I wanted to make some fast cash a few weeks ago. I'm a web dev with a decent amount of experience, so I figured I'd build something small for a local business and sell it. The catch: I didn't write most of it. Claude Code did. I described the idea and it produced a working SvelteKit demo in about 40 minutes. I deployed it to my own server and gave each coffee shop its own subdomain, and the demo loaded with their logo and name already on it. Then I walked into three shops near my apartment with something they could tap on instead of a pitch deck. The first owner said yes in five minutes. $700. Since this is ai_agents channel, I'll be straight: the thing I sold isn't an agent. It's a normal web app. The agent in this story is Claude Code, and it did almost all the engineering while I handled the parts it can't, like walking into a shop and reading whether the owner wants this. Every table has a QR code. A guest scans it, the app reads the table number from the code, and they order from their phone. The order shows up in a barista CRM with the table number and items, so nobody waits for a waiter to write it down. Staff get their own logins too, which means a waiter can work five tables in one lap and push each order to the bar instead of walking back to the register every time. The owner cared most about loyalty. A customer logs in with Telegram, places five orders, and keeps a 20% discount after that. Telegram is the main messenger where I live, and it lets you wrap a web app as a mini app, so I shipped that version too. The discount isn't the point. The shop now owns a customer list and can message those people on their phones. Someone has lunch, joins the program, goes home, and the next morning gets "two lattes for one today" as a notification. A PDF menu doesn't do that. I haven't seen another shop in this city running anything close. Core build took three days through Claude Code. I spent about another week on fixes and sign-offs, and most of that was me waiting on the owner to reply, not writing code. It's been in production for a while now, serving real customers every day and sending me logs and monitoring. Stable so far. The $700 isn't the interesting number. The ratio is: a few hours of agent work plus a walk around the block produced a deployed, paid product. Most of my time went to finding the buyer and keeping it running. I also got a permanent 50% discount at the shop, which doesn't hurt. The bottleneck moved off the build. A question for the people doing the same thing. If you sell these apps to small businesses, do you get a long tail of bug reports coming back at you? I get almost none, but I've been building web apps and shipping products for years, so maybe that's the reason. I'm curious about the people who never wrote code by hand and jumped straight into vibe coding. Does it hold up for them, or does the tail show up?