I don’t write tickets or look up parts anymore. I take a picture of the model number and the tech sheet, tell it what the customer said, and read what it writes before it goes on the job.
This is the step past chapter 1. In chapter 1 Claude writes the two line items and you paste them into your ticket. In this chapter it has a key to your CRM, so when you say “write up the ticket,” it writes the ticket. The labor line with the whole repair in plain words, the part line with the number and the price, and a note on the job with the model, the serial and what you found. You read it, you say yes, it’s on the job.
The second half is memory, and the order matters. I photograph the model number first. If it already has the tech sheet for that model, we go. If it doesn’t, I photograph the document, whatever the maker calls it: the tech sheet, the mini manual folded inside the door or behind the kick plate, the wiring diagram, or the full service manual, or I download it from the makers who give them out, and it gets filed by model with the fault codes, the service mode, the test points and the wiring typed out. If there’s no sheet to be found, it still helps, but noticeably less. The document is what makes it good. With the document in front of it, it’s a tech line: it has the sheet, it has what failed on that model last time, and it never puts you on hold. Whether it beats the manufacturer’s tech line is my opinion, not a measurement, and my opinion is yes.
This morning that was a GE dishwasher with an intermittent dead display: hold the left pad and Start five seconds, count the blinks, eight blinks is a stuck key, and the service light on the main board under the tub tells you whether it’s the panel or the board. That’s off the sheet, in plain words, before I had the door apart.
It will not fix the appliance for you. What it does is get a tech close, and tell them what to check next, the way you would if a green tech called you from the customer’s kitchen and you were talking them through it. Status, honestly: the tickets and the tech-sheet filing are in use, every day. The memory across visits is a month old and growing. A new hire has not run on it yet, because I’m one truck. That last part is what I expect, not what I’ve measured.
As of September 2026 this needs a paid Claude plan, because it runs in Claude Code, which the free plan doesn’t include: Pro at $20 a month, or Max from $100 if you use it all day. Your CRM has to offer API access, and some only do on their higher plans; on Housecall Pro it’s a setting on the higher tiers, and I won’t quote which, check your own plan page. Time cost: an afternoon to set up, then it saves you the ticket-writing on every job after that.
If you don’t have a CRM. You don’t need one to start this chapter. The shop folder, the RULES file, the tech sheets and the ledger work exactly the same with no CRM at all; the only step that changes is where the finished ticket lands. Claude’s Google Calendar connector can put the job on a calendar it can see, and on a Mac, Claude Code can write to Apple Calendar and Apple Reminders. I run Housecall Pro, so that isn’t what runs in my shop; it’s what the tools do, and it’s where I’d start if I weren’t ready to pay for a CRM. Move to one when you’ve got a second tech, or when you’re tired of typing invoices by hand.
The five percent. If you have a CRM and you don’t like it, you don’t have to keep using the parts you don’t like. If you don’t have one and you want one, buy it for the five percent you’ll actually use and let the system run that five percent. Nobody uses all of a CRM, and everybody pays for all of it. My guess, and it’s a guess from years of paying for one, is that most shops use five to ten percent of what they bought, and every shop uses a different five to ten percent. That’s why nobody can build the right software for you until they know which five percent you use, and that’s the one thing you know better than any software company. Now you can build your own five percent. Write down the five screens you actually open in a week. That list is the spec, the RULES file is the rest, and the AI does the building. I’m testing that on my own shop right now, and it isn’t running my business yet, so this paragraph is a belief and a plan, not a report.
The chat app and Claude Code are two different tools, and I use them for two different jobs. The chat is where I think. Claude Code is where things get built and changed. The chat can’t touch anything, so it’s the safe place to be half-right out loud. Claude Code has its hands in your files and, in this chapter, in your CRM, so by the time you open it you want to know exactly what you’re asking for.
So the habit is: bounce the idea off the chat, then have the chat write the brief.
Write a brief I can hand to Claude Code to build this for my shop. One page. Use these headings: What I want. What done looks like. What you must not touch. What I already have. Questions to ask me before starting. Plain words, no code. If anything I've said contradicts itself, say so at the top.
I go one step further. I hand the same brief to a second chat, sometimes one from a different company, and tell it to tear the brief apart before Claude Code ever sees it. Two models that never saw each other’s work agreeing on something is worth a little. One model agreeing with itself is worth nothing.
Keep it to ticket writing and tech sheets for the first month, with the show-me-first rule in writing, and read every line it puts on a job. Don’t hand it the phone, the calendar or the money until you’ve caught it being wrong once and seen how it handled being corrected. And if you’re thinking about a new tech: put them on it the first week, but you read every ticket they let it write. The system gets a green tech close. It doesn’t get them there.