CCAZ · SESSION 1 OF 2 · AI FOUNDATIONS · OCT 27, 2026
AI foundations for the people who build commercial work.
This is a working session for the PMs, estimators, project coordinators, field ops managers, and the office that keeps a concrete company paid. You leave knowing what the tool is, where it is safe to use, and one method you can run on your own pile tomorrow morning. Two goals.
Goal 01
Know what it is, what it is not, and where it is safe to put company information.
It is not a search engine and it is not an expert. It is a very well-read guesser. Once you understand that, the confidentiality rules, the IT policy question, and the "can I trust this" question all get easier to answer.
Goal 02
Leave able to run CRIT on a real pile from your own job.
Most people type a question, take the first answer, and decide the tool is mediocre. That is a search box with extra steps. You are going to watch what changes when you make it ask you questions before it answers, and then you are going to do it yourself.
Before we dive in
A little about me.
I'm Brennan Gerle. I help organizations look past the hype of artificial intelligence and build practical systems that pay off in real work. I founded POLR AI in Chandler because businesses knew they needed AI and nobody was showing them how to put it in without losing the human touch they had spent years building.
My work is not really about automation. It is about how the business actually runs. I sit with leadership first to find where the operation is stalling and where a system would actually free capacity. Then I coach the people doing the job so they feel equipped rather than replaced, and we develop the systems inside the work they already do. Most of my clients are in construction, the trades, and manufacturing. Companies that run crews and pump trucks in the morning and chase a GC for a change order in the afternoon. That is this room.
Consulting
Leadership sessions and workflow discovery come first. We rank the pain by dollars and hours, so the first thing we develop is the one that pays for the rest.
Coaching
Sessions like this one. The people doing the job learn the method, not just the tool, so it still works when the software changes.
Developing
Working software inside the systems you already have. Internal tools, workflow automation, role-specific assistants, dashboards. Built with your people so they own it after I leave.
Section 01 / 06 · Room scan
Where is the room?
Fifteen of you answered the survey, and the board below starts with those answers. If you did not fill it out, raise a hand for your level and we will add you live. There is no wrong answer here, and the count decides how fast we move.
The room, starting from the survey
Total: 0 people
15 from the survey. Add hands as they go up.
Never usedYou have not tried a chatbot yet. One of fifteen on the survey.
A few times a monthYou have asked it to draft an email or reword a document. Three of fifteen.
A few times a weekIt is part of the week, not yet part of the day. Five of fifteen.
Every dayIt is already part of the work. Six of fifteen, and four of you use the AI built into Procore, Sage, or Bluebeam.
What you told me on the survey.
Fifteen of you answered. That is not the whole room, but it is enough to set the pace and pick the examples.
From the survey · where you are
6
of 15 use it every day. Not a beginner room.
Every day 6A couple times a week 5A couple times a month 3Never 1
ChatGPT 14/15Copilot 6Gemini 5Procore, Sage, or Bluebeam 4
They already use the tools. They have not been shown a method.
From the survey · what is eating the week
5
named change orders, impact letters, RFI narratives
Billing: installed against contracted quantities then
From the survey · the hours
~230
hours / week total. 14 of 15 put a number. Median ~11.
Estimating60
Change order cluster, 4 people40–45
Email and takeoffs35
Submittals and insurance certs25
Documentation and rewording25
From the survey · what worries you
5
accuracy, in five different wordings
Accuracy: precise for the trade, mistakes treated as fact 5
Left behind / want to know how to use it 4
The crutch, including leadership: lean on it without the why 3
Section four is built from these answers.
Fifteen replies. Percentages of fifteen, not of the industry. Hands above are the count for today.
Section 02 / 06 · What it is, what it isn't
It is not a search engine and it is not an expert. It predicts what comes next.
That is the whole mechanism. It read an enormous amount and got very good at guessing which word, number, or line should follow. None of what it read is about your job, your owner, or your subs. That is why it can sound completely certain and be completely wrong, and it is why a person stays at the end of every one of these.
Click any card to flip →
Sixty seconds on safe use
Do not paste the GC's contract documents, your unit pricing, bid numbers, or employee information into a free chatbot. If your company has an IT policy or a company account, use it. Today we use a labeled sample, not your project.
Where does the information go?
Three questions to ask before a real document goes in: who can see it, does it train the model, and does my company already have an account.
The free version is fine to learn on. Once real project information is in the chat, you want a company or paid account with training turned off. Your subcontract almost certainly has confidentiality language about the owner's documents, and your unit pricing is the most valuable thing your company owns. Pasting either into a free chatbot is the same category as forwarding them to a personal Gmail. Click a platform for the "turn off training" walkthrough.
How to turn off training, by platform · click to expand
Getting good at this has five layers. Today is the first and beginning the second.
Almost everyone stops at layer one, decides the tool is mediocre, and quits. That is the single most common reason somebody tells me AI did not work for them. Today we do layer one right and begin layer two. Session 2 continues with layers two and three.
Today
01
Prompt engineering
Writing the ask so one good answer comes back. Who you are, who it should be, what you want, and when to stop.
This is CRIT, next section
02
Context engineering
Deciding what it gets to look at before it answers. A perfect question against the wrong pile is still a wrong answer. Most bad results are this, not the wording.
Session 2
03
Harness engineering
The setup around it. What it may read, what it may change, and where it has to stop and wait for a person. This is the layer that keeps you out of trouble.
Session 2 · connectors and skills
Past both sessions, so you know the words exist
04
Loop engineering
One assistant running the same cycle on its own: do the thing, check the result, fix it, go again. The hard part is teaching it how to know when it is actually done.
Runs without you watching
05
Graph engineering
Several assistants with defined jobs handing work to each other, the way a crew does. Who owns what, who checks whom, and what gets called off when the facts change mid-job.
An org chart, not a chatbot
Section 03 / 06 · How you talk to it
Most people ask AI for an answer. CRIT asks it to think with you first.
CRIT is a framework from Geoff Woods, who wrote The AI-Driven Leader. Four parts, in order. It is the difference between a tool that hands you something generic and a thought partner that admits what it does not know before it opens its mouth.
C · Context
The background it does not have. What job you are on, what the foreman found, what the GC said, what is at stake, what you already tried. Skip this and it answers a question nobody asked.
R · Role
Who you need it to be. A concrete PM who has written a hundred change order requests and knows a verbal from a GC super is worth nothing on Friday. An estimator who has been burned by a rebar takeoff. The role decides which questions it thinks to ask.
I · Interview
The part everyone skips, and the reason this works. You tell it to ask you questions before it answers. Same sentence every time: interview me one question at a time, up to three questions, to gain more context.
T · Task
What it produces once it has interviewed you. Format, deliverable, and the stop. Draft it, do not send it. The stop is part of the task, not an afterthought.
The Google method versus the CRIT method.
Both of these can give you a decent answer. The difference is how many decisions the thing made for you before you saw it.
The Google method
you type a question→it guesses→you catch the misses
You end up the reviewer. The work is checking someone else's assumptions about your own job, and you only find the bad one if you look hard.
The CRIT method
you set it up→it asks back→you decide
You stay the one steering. It cannot guess wrong about the thing it stopped to ask you, and nothing gets sent, filed, or issued without you.
The part that stays yours.
You will hear that this makes people lazy, or that it is a crutch for someone who cannot do the job. Here is the honest answer. A good prompt is not a way to think less. Every one of those four letters is a place where you put something in that nothing else has.
The context is yours
It has read almost everything ever written and none of it is about your job. It does not know your GC, which trade is in front of you, or what the super said at the trailer. Every answer worth having starts with something only you could have told it.
The judgment is yours
It drafts. You decide what goes to the GC, what goes to the engineer, and what waits until you walk the pour. That is not a safety rule bolted on the end. It is the whole design.
The name on it is yours
When that change order request reaches the GC or that RFI reaches the engineer, nobody is going to ask what the model thought. Twenty years of a GC trusting your paperwork is not in the training data.
Which is the honest case for the Interview line. It is not a trick to get better writing out of the thing. It is the step that makes it come get what is in your head before it guesses at it. A person who runs CRIT well is doing more thinking about the job, not less.
Same messy pile. Three different asks.
Toggle Lazy, then Better, then CRIT on the sample below. This is four minutes. Then you build one of your own.
Worked examplesSample data only, one made-up tilt-up job where we are the concrete sub. The answers below are worked examples written for this session. Run the CRIT version live on the day so the room sees the real thing ask its first question. What changes across the three is not the quality. It is how many decisions the model makes for you without asking.
You
AI
What changed:
Now build one of your own.
Pick a pile from your own week. Type into each box. The fields turn green as they get specific enough. Copy the prompt. Paste it into ChatGPT, Claude, or Copilot. Keep the GC's name and the project name out of it for today.
CContextEmpty
Who you are, what landed, what's at stake. The job, the phase, the mess.
Type your situation. Real owner and project names stay out of the chat.
RRoleEmpty
Who you want the AI to be. Years, specialty, the lens to bring.
Tell the AI exactly who to be at the desk.
IInterview
Locked
This is the secret ingredient. Locked in so you don't forget.
This line tells the AI to ask up to three questions, one at a time, before answering. We've locked it so you don't skip it.
TTaskEmpty
What it should produce after the interview. Format. Deliverables. And the stop.
Tell it exactly what to make. Include the stop.
0% · Start with Context
Your CRIT prompt
Fill in the fields above. Your assembled CRIT prompt will appear here, ready to paste into ChatGPT, Claude, or your platform of choice.
Live response
First example we will run · one morning as the concrete sub
This is the first example we will run.
CRIT (the four letters) is how you build the prompt in the moment instead of hunting a saved one. A year ago the advice was to collect hundreds of prompts and paste the right one. That binder is already out of date. If you can describe the pile, name the desk, make it ask, and say what and when to stop, you can write it for what is in front of you, faster than you could find the saved one.
Step 01
Describe the pile
Two or three sentences. What you run, what landed, what is at stake.
Step 02
Name the desk
Who would you hand this to if you had them on staff? That is the role.
Step 03
Make it ask
The locked line. Three questions, one at a time. Answer them like you would answer a new hire.
Step 04
Say what and when to stop
A brief, a table, a draft reply. Then the stop. Nothing goes out without you.
Section 04 / 06 · Trusting the result
Five of you said accuracy. Here is how you decide whether it is right.
On the survey the worry was not "will it work." It was "mistakes," "precise and accurate for our trade," "it still gets things wrong and people use it as fact," and "saving time that creates expensive mistakes." Fair, and correct. Here are four habits that answer it. Each one takes less time than the retyping it replaces.
01Point
Point at the source
If it cannot name the email or sheet a number came from, that number stays off the change order request.
02Show
Show the math or the source
Ask for the arithmetic or the source document, not the answer. A total you can check in ten seconds beats a confident one you cannot.
03Check
Check it against one thing you know
Read the part you already know first. If that is wrong, stop and fix the context, not the answer.
04Ask
Let it ask first
A model that asked you three questions has fewer places to guess. Most hallucination is a gap nobody told it to fill.
Same checks you run on a new project coordinator's first change order request. You do not need to know how the model works.
Where is it reliable enough? One of you asked exactly that.
The owner's survey answer was the best question in the set: where is AI reliable enough to automate, where does it still need a person, and how do you build safeguards so saving time does not create an expensive mistake. Here is the honest version of the answer, as a ladder.
Reliable today · read and brief
Reading a pile and telling you what is in it.
Email, the foreman's texts, a stack of daily reports, an RFI response, an insurance cert against the GC's requirement. If the brief is wrong, you lost two minutes and you can see why. Every task on your survey starts here, and this is where everyone should be tomorrow.
Reliable with review · draft
Drafting the change order request, the RFI narrative, the impact letter.
The words are usually good. The numbers are only as good as the documents you gave it and the questions it asked. Nothing sends. You read it the way you read a project engineer's draft, then you send it under your name.
Not yet · act on its own
Takeoffs you did not check. Anything that posts, sends, or files itself.
A rebar or yardage takeoff is a quantity with your name on a bid. If you cannot show the math, it is not reliable enough, no matter how fast it was. Letting it act on its own is Session 2, with a connection you control and a stop you set. Nobody should be here after one morning.
Three of you said "crutch." You are right to.
From the survey
"It becomes the guide to be followed without a person understanding the why and how."
The Google method does exactly this. You type, it answers, you stop. CRIT runs the other way. You have to describe the job, name the desk, and answer three questions about your own project before it writes a word. That is more thinking about the work, not less. If your people are using it the first way, the fix is the method, not a ban.
From the survey
"Relying on AI for a takeoff and not actually understanding the project."
Correct, and it is why takeoffs are on the "not yet" rung above. Use it to read the spec, pull the mix design requirements, list the assumptions, and check your count against the drawings. Do not use it to produce a yardage you could not defend to the GC. The estimator who understands the project is the one who can tell when the tool is wrong.
From the survey
"People will become complacent and not back-check it."
The four habits above are the back-check, and they take less time than the retyping did. Make them the standard the same way you made a second set of eyes on a pour ticket the standard. The tool rewards the people who already know the work, because they are the ones who can check it.
Also from the survey
"The responses are super long and could be shortened."
Yes. That is the Task line. Tell it the format and the length: "one page," "five bullets," "a table with these columns." It will do what you say. A long answer is usually a prompt that did not say what done looks like.
Section 05 / 06 · Live: CRIT on what you named
You named change orders, letters, and tracking. Let's run them.
Change order requests, impact letters, and RFI narratives were the most-named tasks on the survey. Tracking was next: company-wide status, installed against contracted quantities, submittals and insurance certs. Same made-up tilt-up job for all of them, and we are the concrete sub on every tab. We paste the sample, run CRIT live, answer its questions as a room, and read what comes back against the four habits.
Live drawing · where the email actually goesCollapsed · pull open to teach
Beat 1. Something lands in Outlook or Teams. That is inbound. You have not touched it yet.
Inbound · email / Teams
YOU
Outlook / TeamsWhere it lands
CRMWho you already know
Procore / ProjectCO log, RFIs, the job
Excel / Word / OfficeThe working paper
QuickBooks / ERPCost, billing
SharePoint / OneDriveWhere the file lives
You sit in the middle of the tools you already use. Click the board or Next beat to walk the three moves. Beat 2 is dump it into Procore or Excel. Beat 3 is send something back and file it.
Fake sample · one morning as the concrete sub on a tilt-up warehouse
Paste this live for the change order request and morning brief runs. The other tabs carry their own sample inside the prompt.
SAMPLE ONLY. Invented for today. Not a real project, GC, or sub.
Project: Building 3, 140,000 SF tilt-up warehouse. We are the concrete sub: footings,
grade beams, slab on grade in four pours, tilt panels. Today is Tue 10/27.
1. From: our foreman (text) 6:12a
"super had us add 2 footings at C/7 and C/8 yesterday, said engineer wants them.
we poured them with the grade beam. about 6 yds extra plus rebar. he said hes
good for it"
2. From: GC project manager 6:48a Subj: RFI 044 response
Engineer's response attached. Grade beam reinforcing along line C revised from
#5 at 12" to #6 at 8" o.c. Please proceed. No change to contract time anticipated.
3. From: GC scheduler 7:05a Subj: Pour 3 date
Slab pour 3 (east half) moves from Thu 10/29 to Mon 11/2. Steel erector needs
the deck clear. Please confirm pump and crew for Monday.
4. From: ready mix dispatch 7:20a Subj: Thu 10/29 order
Confirming 180 CY, 4000 psi, 3 trucks rotating, first load 5:30a Thu. Orders
cancelled inside 24 hrs are charged per the account terms.
5. From: GC accounting 7:41a Subj: Pay App 5 approved
Approved at $212,400 against $248,900 billed. Deducted 4,100 SF flatwork "not
in place" per super walk 10/22, and $6,800 cleanup backcharge. Check cuts Friday.
6. From: our estimator 7:55a Subj: Friday bid
Need rebar tonnage and CY from the structural set for the Friday bid by Wed EOD.
Can you or the coordinator pull it?
7. From: safety 8:05a Subj: Toolbox talk
Reminder: toolbox talk Wed 6:30a, topic is silica. Sign-in sheet to me by noon.
No real names. If the foreman says "he is good for it" or the GC moves a pour without moving your concrete order, that is on purpose. The flags are the lesson.
10:00 to 10:30 · Build a workflow together
Five of you write the same document. Let's build the Change Order Request workflow and take it home.
Change order requests, delay and impact letters, RFI narratives, documentation and rewording. Five survey answers, one job: turn something the foreman said or the GC did into paper that gets you paid. We build it as a saved CRIT prompt with a fixed output, so a PM, a coordinator, or a field ops manager runs it the same way every time. It reads and drafts. It does not send. That is the level of trust this room agreed on in section four.
Get
The claim, however it arrived
The foreman's text, an RFI response that changed scope, a pour the GC moved, a pay app that came back short. Paste it as is. Add the one subcontract clause that governs it if you have it.
Move
Four outputs, same order every time
A confirming email that puts the verbal in writing today. A change order request with blanks where there is no number. A CO log line. An impact letter or an RFI narrative, only if one is owed.
Decide
You read the blanks first
Every number points at the document it came from or it is a blank. Send the confirming email now, the pricing when you have it. You send it under your name.
Save
Run it on your next one
Same prompt, new claim. Bring five of them to Session 2 and we connect it to the CO log, the schedule, and the accounting system.
Your turn. What did a GC or a foreman hand you this week that should have been in writing? Call it out. We will run CRIT on it if we have time.
None of this replaces your judgment. It gets you to a working draft fast. You still review, edit, and own what goes out.
Section 06 / 06 · Tomorrow morning
Pick one pile. Run it tomorrow.
Do not try to change how the whole company works on Wednesday. Take your next change order request, or one morning's texts and emails, or one short pay app. Strip the GC's name if you need to. Run CRIT. Read the blanks first.
Tomorrow morning
Take the next thing a GC or a foreman hands you that should be in writing
Strip the GC's name and anything under the owner's confidentiality clause
Run the change order request workflow. Let it interview you. Answer honestly.
What "done" looks like
A confirming email you could send in ninety seconds
A change order request with every number pointing at a ticket, a text, or a drawing, and blanks where it does not
A flag list you actually look at before you answer the GC
What you skip
No unit pricing or bid numbers into a free chatbot on the jobsite Wi-Fi "just to try"
No sending anything it wrote without reading it
No "the model can just do the takeoff." That is Session 2, with rules.
Before Session 2
Run the workflow on your next five claims. Notice what it gets wrong and why.
Write down one task you or your team does every week that starts with a pile
Find out whether your company account can connect to email, your CO log, and the GC's system you live in
What Session 2 covers.
Session 1 was the method. Session 2 starts with an exercise: identifying the valuable workflows in your week, the ones that repeat, cost time, and follow a pattern worth building on. From there we turn one into a skill, a saved way of working the assistant runs the same way every time, and we spend real time on validating and trusting your results: what to check before you send anything, and where the model is reliable and where it is not. Then we talk about connectors, the plug into your email, your schedule, your accounting system, and the GC's project software you already live in, and the two permissions you control on every one of them. Estimators, anyone who does rebar and yardage takeoffs from a structural set, this is your session.
Common questions
You're probably wondering one of these.
Click a question. These are the ones that come up every time.
Click a question to see the answer.
Open the floor
This is the part that matters. Your inbox. Your submittal log. Your "we already tried ChatGPT and it was meh." Let's go.
Next step
Run it for a week. Bring what happened to Session 2.
If you want help before then, bring the actual pile, with the owner's name already stripped.
Scan this
Everything from today, plus the sample files and the prompts, on one page built for this group.
