Somebody Has to Walk the Site

There are two stories going around about this technology, and if you build for a living you will meet both long before you meet the thing itself.

The first one ends the trade. Houses printed over a weekend, a robot arm laying block dead straight in the sun, a drawing set that turns itself into a building with nobody on site. There is always a video. It is always shot somewhere flat and dry, with good soil and no inspector in the frame. It gets watched the way people watch a storm forecast for the next county over: a little alarming, mostly entertainment.

The second story is on the show floor every winter, with a booth and a demo loop. That one bids your jobs, builds your schedule, chases your subs and finds you a framing crew by March. Thirty days free. There is somebody at the booth who will explain it in the voice of a man doing you a favor.

Both stories run on the same fuel. They are huge, they are set in a year nobody has worked through yet, and neither one gives you a thing to do at six thirty tomorrow morning. Believe whichever you prefer. The week itself does not move. The basement bid you promised by Friday is still not written. Your clients still have not picked tile, so the setter you booked for the eleventh is now a problem. The rot you found in the rim joist on Tuesday is still not on paper, and every day it stays off paper it gets harder to bill.

Nothing in this book is a prediction. All of it fits inside the week you are already in.

The first story does not survive a site visit.

It cannot walk the site. It cannot stand in the driveway of a 1962 ranch on a sloped lot and know, before anybody opens the door, roughly what is behind that drywall and what finding out will cost. It cannot lift the crawlspace hatch and smell that there is water down there and has been for years. It cannot look at what got framed this week and hold it against sheet A-3 in its head, which is how you catch a rough opening an inch and a half narrow eight weeks before the windows land. It cannot tell whether that slab is going to finish flat. It cannot tell from four seconds of a phone call whether the drywall sub saying Monday means Monday or the Monday after.

It cannot stand in front of an inspector either. When the framing inspection comes back with a correction on it, that conversation is between a person with authority and a person with a license, on a site, in front of the actual work. No software is in the room and none is coming. Your license number is on the permit. Your name is on the contract, the lien waiver and the certificate of insurance, and none of those has a line where a chat window could sign.

It does not know the rules where you work, and it will sound like it does. Ask about a span, a fastening schedule, the deposit your state lets you take up front or the cancellation notice your contract has to carry, and back comes a clean confident paragraph describing some other jurisdiction’s rules, or nobody’s. Codes get adopted by edition and amended locally; home-improvement contract law is written state by state. Chapter 10 deals with it properly. Until then, treat what the tool says about code and contracts the way you treat a lumber price somebody overheard at the counter.

So the first story is a no, and it did not need much explaining. Not the building, not the judgment, not the license, not any document that carries legal weight. Anyone who has to spend a chapter convincing you your trade is safe is working up to a sale.

That is the reassuring half. Reassurance on its own is worth nothing, so here is the other half: something in this business is changing, and it is nowhere near the framing.

The first piece of it turns up in the kitchen, during the walkthrough. The homeowners have already asked a machine. They arrive with a number, a scope, a materials list and eight questions, all of it produced, confidently, by something that has never seen their house or their permit office. Some of it is reasonable. Some of it is nonsense priced for a different state. Arguing with it does not work. What works is being the company whose paperwork is clearer than the machine’s: a scope written in their words, an exclusions page that shows where the edges are, an allowance section that reads like honesty instead of a trap.

The second piece is simpler and sharper. The outfit bidding against you can now put a real proposal in that inbox inside forty-eight hours. Not because anybody over there got better at writing. Because writing got cheap. A homeowner choosing between three companies for a ninety-thousand-dollar job cannot judge anybody’s framing and knows it, so they judge the paper, which is the only evidence they have. That contest is not new. What is new is the price of entering it.

Which is the actual point of this book. Nobody mentions, on the day you file the paperwork for your own company, how much of general contracting is typing. Every job you have run produced two things: a building and a stack of paper, and only one of them gets photographed. You write the scope, the exclusions, the allowance schedule and the cover letter that almost nobody you bid against will send. You write the change order for the sink they asked you to move, or you build it and forget it. You write the weekly update, the delay message, the decision deadline on tile, the sub confirmation, the RFI to the architect. You write the draw request, the progress invoice, the friendly reminder, then the firmer one. You mean to write the daily log you know wins disputes, and at seven fifteen the legal pad loses to the couch.

None of that is building, none of it is billable, and it all lands in the same place: after the last truck leaves, or it never happens at all.

The stack is what these tools are actually for. Not deciding what a job needs. Writing up what you decided, in plain sentences, about as fast as you can talk.

The tool is bad at things too, and the demo loop will not mention them. The main one is that it makes things up. Not sheepishly, and not in a different font: the invented material arrives in the same steady voice as the true material. Ask what a kitchen runs where you build and you get a figure produced out of nothing. It has no idea what your electrician bid last week, what the yard charged you this morning, or where your markup sits.

So the workflows in here are built the other way round. The numbers stay with you — takeoffs, sub quotes, markup, dates, durations. The sentences come from the tool. Everywhere that line gets thin, a WATCH OUT box says what to look at and how long looking takes. About a minute, usually. That minute is what the tool costs you. Set it against the evening it gives back.

Here is a week that anyone who runs their own jobs will recognize. The strange part is that nothing in it goes wrong.

Tuesday at four thirty you walk a basement finish. Good people, decent house, a job you want. You had the shape of the number before you were back on the highway, and you told them end of the week. Friday goes to a framing inspection that does not go the way it should and a truss delivery two bundles short. The proposal goes out eleven days later, typed after nine at the kitchen table, mostly boilerplate with a price at the bottom. By then they have signed with the company whose proposal landed the following Thursday, with a cover letter that repeated what they said in their own kitchen and an exclusions page that made them feel like adults. You put it down to price. It was not price.

Same week, Wednesday, you are on the deck job and the client says while you are at it, could we move the hose bib and add two cans in the hallway. You say sure, because it is half a morning and they have been easy to work for. Nobody writes anything down. At final billing you cannot reconstruct it honestly, and you will not pick a fight at the end of a job you are proud of, so you eat it and call it goodwill.

Neither of those feels like a loss on the day it happens. Nobody was lazy. Nothing was mismanaged. They show up at the end of the year, as the difference between a decent year and a good one.

And nothing in that week was a skill problem. You know how to write a proposal that wins at the top of the range, because you wrote one on a Sunday that ran nine pages and won. The trouble is the clock. The good version costs an evening, and the week hands you three to write and one evening to write them in.

So here is the book in plain terms.

This is a manual for putting an ordinary chat tool to work on the written half of running construction. Proposals, scopes, exclusions, allowance explanations, cover letters. Change orders written the same day the client asks. Weekly updates, delay news, decision deadlines, sub confirmations, RFIs. Draw requests, progress invoices, collections, supplier RFQs. Review replies. Daily logs, punch lists, crew briefs, toolbox talks. The marketing that has been on the list since last spring. All 150 prompts are numbered and gathered again in a library at the back, which is how you find the right one while somebody stands in your driveway waiting on you.

The prompts were run through ChatGPT, Claude and Gemini before they went in, and anything that only behaved on one of the three did not make the book. None of it is a guess at what the tools might manage for a builder someday. It is what came back usable on all three.

What it is not: an estimating system, a takeoff tool, a code reference, a contract drafter, or anything that gets a vote on safety. It will not price your job, and this book never asks it to. Five things in this business can bite you through a chat window: client data, contract language, lien deadlines, advertising rules and safety. Each one carries a RULES CORNER box where it comes up, and Chapter 10 puts all five in a single chapter. Read that one before any client’s details go near a chat window.

You do not have to be good with computers, and you do not need project management software, though Chapter 11 covers it if you have some. What it takes is a phone, an email account and thirty spare minutes. Chapter 2 walks the setup. Chapter 3 teaches the one skill worth learning, and you have been using it for years: writing a bid package a sub can price without calling you back with three questions is the same act as briefing one of these tools.

Then the size of it, said once, so nobody oversells you. This does not change your life. The Promise Page after this one adds the hours up, about eleven and a half in an ordinary week. The number is real. It is also not free, because the first week costs you more time than it gives back, the way every new tool does. Around week three the sums turn over. What you are buying is the hour after the last truck leaves, and then most of the evenings in the week.

Two habits is what most people keep out of a book like this, and two is enough to pay for it. Usually it is the Friday update and the same-day change order. The marketing chapter never gets opened, four or five hours a week come back, and after a month it stops feeling like using AI at all. That is a complete result, not a half-finished one. Twelve chapters is a menu, not an assignment.

Somebody still has to walk that site. Somebody has to read the slope, price the risk, catch the framing before the drywall covers it and stand in front of the inspector, and that will be a person with hands and a license for a long while yet. What is actually on the table is smaller than either story you were sold and worth more than both: the proposal in their inbox while they still remember you standing in their kitchen, the change order signed in the driveway instead of argued about in February, and a Thursday night that belongs to you.

Over the page is the arithmetic, chapter by chapter.