Article

How to Use AI in Your Contracting Business: Quotes, Admin, and Office Work

July 7, 2026By Bor Cerlini

The Short Version

How to use AI for contractors, in one line: put it on the office work. The quotes, proposals, customer emails, SOPs, and paperwork that keep you at the kitchen table at 9pm are exactly the jobs AI does well.

Most owners have already tried and got junk back. That isn't a sign AI can't do the work... it's what happens when the AI has never seen your jobs, your rates, or your margins, so it fills every gap with an internet average.

The fix is a one-time setup: a permanent, written memory of your shop, in plain files you own, on your own computer. Build that and the same AI that shrugged at you starts doing real work.

This guide covers what AI can actually take off your plate, how Claude, ChatGPT, and Gemini compare for it, the folder setup that builds the memory, and the level above it... one brain your whole team can ask.

What AI can actually take off a contractor's plate

The wins are in the office, not on the roof. Everything in your business made of words and numbers is fair game: pricing support built from your own job history, proposals and scope sheets in your format, customer emails and change orders that land firm and clean, review replies, crew SOPs, job debriefs that actually remember the last debrief.

That's most of the admin that eats your evenings. Drafted in minutes, finished with an edit instead of written from scratch... and none of it requires touching the customer-facing side of your business until you're ready.

So why does it feel like AI can't do any of this when you try it? That's worth understanding before you build anything, because the fix follows straight from the cause.

Why your first attempt gave you generic junk

Most contractors' first serious attempt at AI ends the same way. You type out the job, the scope, maybe your rough numbers, and ask for a price or a proposal. What comes back reads like it was written for a company that doesn't exist... a price pulled from some national average, a scope sheet a first-year could have drafted.

So you fix it by hand, and somewhere around the third rewrite you do the math: the ten minutes it saved you just cost twelve. After enough rounds of that, you quietly file AI under overhyped and go back to pricing jobs in your head.

Here's what actually happened in that chat. The AI has never seen one of your jobs. It doesn't know your labor rate, your markup, what that last bathroom really cost you, or the work you've learned to walk away from. Every one of those gaps gets filled with a guess, and the guessing is where the generic comes from.

And the next chat is worse, because it remembers none of it. You explained your business yesterday, and today it's a stranger again. That's not a dumb tool... that's a sharp one working blind, over and over.

Which means the fix was never a cleverer prompt. The fix is sight.

Contractor at a laptop late at night reading a generic AI answer

Give the AI a memory you own, and everything changes

The operators getting results that look like a different tool entirely all worked out the same quiet thing. Instead of re-explaining the shop every time and hoping the AI holds onto it, they wrote it all down once, in files they own, and set the AI up to read those files before every task.

Think of the binder you'd leave a foreman running the job while you're out of town. You don't expect him to read your mind. You leave him what he needs to run it your way, so he doesn't call you every five minutes. Same idea here: the AI reads your files first, then works the job like it's been in your shop for years.

The ownership part matters more than it sounds. If you slowly teach an AI your business inside some company's app, that knowledge lives on their servers, under their rules. OpenAI retired GPT-4o earlier this year, and people who had built their whole routine around it woke up to find that work gone. Files on your own computer can't be retired, watered down, or taken away... and they move with you to whatever AI wins next year.

One more thing this changes: the writing. An AI that has samples of your real proposals and emails stops sounding like a robot and starts sounding like you, which matters the moment a homeowner reads it.

ChatGPT vs Claude vs Gemini: which AI should a contractor use?

For this setup, the comparison comes down to one capability almost nobody talks about: can the AI work inside a folder on your computer, reading and updating your files as it goes? That folder access is the engine of the whole system. Smarts matter less than you'd think... all three are plenty smart.

Claude is the one to use right now. The desktop app works directly with a local folder, it holds the most context at once (your whole business plus the job in front of it), and it writes the cleanest by default... fewer of the AI tells your customers are starting to spot. It's also the most careful with numbers, which is the trait that matters when the file it's reading is your pricing.

ChatGPT is fine to keep for quick questions, but its desktop app still can't live inside a folder on your computer the way this system needs. OpenAI's answer is a separate tool called Codex, which can work with the files and folders on your machine. It's real and it works... it's also built for developers, and it feels like it. If ChatGPT is your daily AI, keep it, and run this setup in Claude alongside it anyway: the owner experience is smoother, and since the folder is just files, whatever you build moves with you either way.

Gemini on its own has the same gap, though Google's free Antigravity tool can give it folder access. Pick that route only if your whole business already runs on Google's stack. Most shops won't be in that camp.

Whichever you pick, the point stands: the memory lives in your files, not in the vendor. You're choosing an engine, not marrying one.

The folder setup that gives AI a real memory of your shop

Here's the actual build, and it's less technical than setting up a new phone. You create one folder on your computer and fill it with a handful of plain text files. No database, no code, no subscription. Something like this:

MyShop-AI/
  instructions.md   <- the AI reads this first, every chat
  shop/
    profile.md      <- trade, service area, team, how you scope
    pricing.md      <- labor rates, markups, margins, terms
    jobs.md         <- past jobs with the real numbers
    voice.md        <- sample proposals and emails, banned phrases
    standards.md    <- your quality bar, the jobs that go sideways

The instructions file is the job description for your new assistant. In plain English: who you are, what's in each file, and the rules it never breaks (always read pricing.md before quoting anything, never invent a number, flag anything outside your service area). Without it, the AI sees a folder full of files. With it, every chat starts already knowing your shop.

The shop files are where the memory lives, and jobs.md is the crown jewel. Every past job you log with its real numbers makes the next answer sharper, because now the AI prices from what your work actually costs instead of what the internet thinks it should.

Then the habit that makes it compound: work inside the folder, and at the end of a session, have the AI update the files with what it learned. Filled in a new job's final numbers? Into jobs.md. Corrected its wording twice? Into voice.md. Six months of that and you have something no competitor can copy... a written brain of your business that gets smarter every week.

Laptop showing a simple folder of business files feeding an AI chat

What changes when the AI knows your numbers

The difference shows up on the first real task. Ask a blank ChatGPT what to charge for a kitchen remodel and you'll get "typically $25,000 to $75,000 depending on scope"... a shrug with confidence. Ask an AI sitting on your last eleven kitchens and you get a worst, middle, and best number built from your own history, with a note that the last one ran over on cabinets.

You still make the call. The AI's job is to hand you your own data, organized, so the number you put your name on is grounded in what your jobs actually cost. That's the difference between decision support and a guess.

And every job from the list at the top of this post gets the same upgrade, because they all draw on the same memory. The proposal reads like you wrote it. The change order quotes the right terms. The debrief builds on the last one instead of starting blank.

The next level: one brain your whole team can ask

Everything above lives on your computer, which is both its strength and its ceiling. The memory is yours, but so is the keyboard. The moment your office manager needs the same answers, the setup has to grow.

That's where a knowledge-base agent comes in: the same idea, moved to where your team already talks. The shop's knowledge (pricing logic, SOPs, the answers to the questions your office asks you ten times a week) gets loaded into an agent that lives in a team chat like Slack. Anyone on the team asks it directly, it answers from your own material, and you stop being the only person in the company who knows how the company works.

If your office has never worked with AI before, roll it out gently... we wrote a separate guide on introducing AI to your team without spooking anyone.

The bigger point is the order of operations. Don't start by shopping for tools. Start by writing down what your business knows, in files you own... every level after that, from a smarter chat to a team-wide agent, is built on the same foundation.

And if you'd rather see this mapped onto your own business first, that's what our free mini-audit is for: we look at how your shop runs, show you what an AI setup or agent could realistically take over, demo it live, and point out where the fastest wins are hiding.

Teach it your shop once, never twice,
Bor Cerlini

Frequently asked questions

Which AI is best for contractors right now?

Claude, because its desktop app can work directly inside a folder on your computer, holds the most business context at once, and is the most careful with numbers. ChatGPT and Gemini both have folder-capable routes (OpenAI's Codex and Google's Antigravity), but both are developer-oriented tools, and neither is as smooth for an owner as Claude's desktop app.

Can I build this with ChatGPT?

You can: OpenAI's Codex tool works with local folders, so the same architecture runs on it. Just know it's built for developers and feels more technical than an owner needs day to day, which is why we still recommend building the setup in Claude and keeping ChatGPT for quick questions. The folder is just plain text files either way, so nothing locks you in... you can switch engines later and point them at the same folder.

What information should I give the AI first?

Start with four files: your business profile (trade, area, team, how you scope), your pricing (rates, markups, margins, terms), your past jobs with real numbers, and samples of how you write. Those four cover most of what makes your answers generic today. Add SOPs and standards as you go.

Is it safe to put my business numbers in an AI?

Treat it like any business system: keep customer payment details and anything legally sensitive out of the folder, and turn off model training in the AI's settings so your material isn't used to train the vendor's models. The files themselves sit on your own computer, which is more control than most cloud tools give you.

Do I need to be technical to set this up?

No. The whole system is one folder and a handful of plain text files you could open in Notepad. If you can write an email describing how your shop works, you can build this. The only discipline that matters is keeping the files honest and current.

When should I move from a personal setup to a team-wide AI agent?

When the answers stop being just for you. If your office manager keeps asking you questions the files could answer, or you catch yourself copy-pasting the AI's output to the team, it's time to move the same knowledge into an agent that lives in your team chat and answers everyone directly.

Written by Bor Cerlini