AI for appliance repair

Try one of these three with a real job this week, and keep your passwords out of it.

This is a plain-English guide to three things I actually use to run an appliance repair shop: a chat program that helps me diagnose, look up parts, price them my way and write the ticket, a phone agent that answers calls and books jobs, and a bookkeeper that watches my QuickBooks. There’s a fourth chapter on what goes wrong and how I catch it, and a fifth on the step past all of it, where it writes the ticket itself and remembers every tech sheet you’ve ever photographed.

Who it’s for

Appliance repair techs and shop owners who have never used AI, don’t use a computer much, and are probably reading this on a phone in a truck. I run one truck. I think everything here holds for a shop of up to three to five techs. Past that, you need somebody whose whole job is running the AI, and that’s a different guide. No programming required for any of it, and I’ll tell you plainly when something does take real technical work.

It’s written for appliance repair because that’s my trade, but the trade is two words on this page. If you’re a plumber, an electrician, a house cleaner, a landscaper, a pressure washer, or you install cameras or mount TVs, read it anyway. If you drive to somebody’s house and they book you on a schedule, chapters 2, 3, 4, 6 and 7 read the same for you, and in chapters 1 and 5 you swap “part number” and “tech sheet” for whatever your trade’s version is.

Never used GitHub, or not sure what “open source” means for you? How this works, and what to click, written for a phone in a truck.

And if you’ve never held a tool for a living, but your desk job looks like it’s going away, read this page first. It’s about you, and about the people who are going to be hiring you.

What this system actually runs, as of September 23, 2026

Not a feature list. A list of things that exist, each with the honest word next to it.

One person, one truck, and a few small subscriptions; chapter 2 names every meter. It’s all because of the AI I’m using and the system I’m building, and this guide is the part of it a shop can start with this week.

The chapters

The blue chapter names below are links. Tap one and that chapter opens. Every chapter ends with a link back to this page and one to the next chapter, so you can read straight through.

  1. Claude on your desktop, the chat program, starting with three small real jobs.
  2. Your phone, answered, a phone agent that answers calls and books onto your calendar.
  3. Your books, connecting a bookkeeper AI straight into QuickBooks Online.
  4. What goes wrong, the habits that catch it when any of the above is confidently wrong.
  5. Your CRM, and the shop’s memory, the step past chapter 1: it writes the ticket, files the tech sheet, and tells a tech what to check next.
  6. One truck to five, the plan, not the practice: outbound calls, hiring, and what’s in use versus planned, dated.
  7. What about robots, the one exception on this site: where I think this is going and what to get ready for, with the facts checked at the source and the belief labeled as belief.

What this is not

Not a course. Not a product. There’s no support line, and nobody’s on call if it breaks for you. It’s MIT licensed and offered as is: you can copy it, change it, and reuse any of it, and nobody is promising it will work exactly the way it worked for me.

Who wrote it

I’m Marc, thirty years fixing appliances, owner of Steel City Appliance Repair in Homewood, Alabama. I’m not a programmer. The phone agent in chapter 2 answers my line every day and the chat program in chapter 1 is on my phone. Chapter 3 is the setup I’d use, written so you can use it on your books; my own year-end still goes to a professional either way.

Why I wrote this

To the guys in the trade groups, most of whom I’ve known for eight years: this is for you. I don’t post much. This is the post.

Here’s what I think is coming, and I’m labeling it the way chapter 4 says to: this is a belief, not a measurement. A lot of white-collar work is going away, faster than most people think. I don’t have a number I’d defend on this page, and I’m testing the numbers I read in a separate series, against the government’s own releases, where I write down when I’m wrong. But the direction I’m sure of. The people who can’t get those jobs are going to come looking for ours, and there is plenty of work. I think it’s on us to be ready to hire them and teach them, and this guide is one shop’s way of getting ready.

Go to college because you enjoy it, not because it gets you a job. Unless you’re there learning to run these systems. That’s the one degree I’d still bet on.

What this site counts, and why you should know it exists

Page views, per chapter, with a tool called GoatCounter. No cookies, no names, nothing stored about you. I use it to see which chapters get read and which ones people leave, and the numbers are public, same ones I see: aiforappliancerepair.goatcounter.com.

I’m telling you because a lot of people don’t know this kind of counting exists, and it runs the other direction too, on your own paperwork. When you email an invoice from QuickBooks with the “online invoice” option turned on, its status moves from Sent to Viewed the moment the customer opens it. An invoice that’s been viewed and sits unpaid for a week is a different phone call from one that was never opened. Housecall Pro does the same when you email from it: it shows sent, delivered, opened, and how many times they opened it. That’s in use in my shop every week; its help pages don’t mention it, which is a reminder that the tool you run knows more than its manual. Nothing tells you which page of a PDF they read. And an “opened” can fire from an email preview nobody actually looked at, so treat it as a hint, not proof.

If something in here is wrong or broke for you

Open an issue at github.com/morriss-group/ai-for-appliance-repair/issues and say what broke, or reply in whatever group you found this link in and say the same thing. I’d rather hear about it than have the next person hit it blind.

Last checked: September 23, 2026.