Talk through your day once
Turn a voice note into a clean daily report
The safest first use of AI: every fact is your own words
The daily report is the paperwork everyone skips and the one piece of paper that wins a dispute. Stand by the truck, talk for two minutes, and the AI sorts your own words into a dated, scannable log with crew, hours, work done, delays, and notes for the office. You supply every fact; it supplies the clean shape; you read it before it counts.
Worth knowing:You are not asking the AI to know anything about your site. It only organizes what came out of your mouth, which is why this is the lowest-risk way to start.
Owners and foremen too wiped to type a report at day's endAnyone who has lost a dispute for lack of a written logCrews that want documentation without more desk time
Start Lesson 1 →Describe the conflict, get the formal question
Write a clear RFI in about a minute
A vague RFI bounces back; a sharp one gets answered same-day
An RFI is the formal question you send when the drawings do not agree, and a vague one costs you days. Describe the conflict in plain words with the sheet numbers, and the AI builds a structured RFI: subject, references, one answerable question, the impact, and a response-needed-by date. You bring the facts and verify every reference; it never decides which drawing is right.
Worth knowing:The AI repeats your sheet numbers verbatim and will not catch a typo. Checking the references before you send is what keeps a sharp RFI from bouncing.
Subs and PMs who dread writing formal RFIsForemen who spotted a plan conflict and need it answered fastAnyone whose RFIs keep coming back as 'please clarify'
Start Lesson 2 →Tailored to today, owned by you
A toolbox talk built for today's actual work
The line that runs through the whole track: drafting aid, not the record
A generic toolbox talk gets tuned out. Give the AI today's tasks, weather, crew, and site conditions, and it drafts a five-minute talk on the hazards your crew actually faces this morning, with plain controls and discussion questions. The boundary is the lesson: the AI drafts, but you read it, deliver it, and keep the signed sign-in sheet, because that sheet is the record.
Worth knowing:Generating a talk does not make you compliant and signs nothing. The draft saves you ten minutes; it does not carry your responsibility. That distinction protects you on every later lesson too.
Foremen and supers who run the morning safety talkCrews that tune out the same laminated cardAnyone who wants tailored safety briefings without the writing time
Start Lesson 3 →Ask the project manual, get the exact words
Find the answer in a 400-page spec book
A twenty-minute hunt becomes a ten-second quoted answer
The answer is in the project manual; the cost is finding it and trusting you read the current division. Upload the manual, ask in plain English, and the AI names the section number and quotes the exact text. It is the index the manual never had. You open the cited section and read it before you order, pour, or submit.
Worth knowing:Make the AI quote the exact text and cite the section. A paraphrase you cannot trace is a rumor; a quote with a section you can open is a usable answer.
Anyone who has lost twenty minutes flipping a project manualEstimators chasing a required strength, finish, or submittalCrews unsure they read the current division
Start Lesson 4 →Ask a 200-sheet plan set where things are
Chat with your blueprints
Build the count fast; verify it true, sheet by sheet
A big plan set hides answers that are drawn, not written. Load it and ask where the wall section is or how many fire-rated doors are on the second floor, and the AI returns a list tied to sheets. The drawing-AI tools say it themselves: it does not replace your review. You verify the list sheet by sheet, especially anything that drives money or safety.
Worth knowing:Checking a list the AI built is far faster and less error-prone than building one cold. That is where the time is saved, and the drawing stays the authority.
Estimators flipping a set to find every instance of one thingPMs and supers chasing a detail callout across sheetsAnyone who has missed a rated door in a count
Start Lesson 5 →Counting and measuring, done first
AI takeoff: upload a drawing, get quantities
An evening of tallying becomes a list you check in a minute
A takeoff is the slow counting-and-measuring part of a bid, usually after the workday, when mistakes creep in. The AI returns counts and measured quantities as an editable list. You start from its numbers and correct them: verify the high-impact quantities by hand, spot-check the rest, and confirm the scale and the scope it could see. The counting hours are the AI's; the total is yours.
Worth knowing:Every measurement rides on the drawing scale, and every count rides on which sheets you uploaded. A missing scale or a missing plumbing sheet gives a confident, quietly wrong answer.
Estimators doing manual takeoffs at night with a scaleOwners who want to check quantities, not tally themAnyone whose bids hinge on getting the counts right
Start Lesson 6 →Auto-accept, spot-verify, or manual-redo
From takeoff to bid, with a trust system
Spend your checking time where the risk actually is
The skeptic's real question is which numbers to trust, and the answer is per-item. Grade every quantity into three tiers: auto-accept the cheap and certain, spot-verify the medium-stakes with a sample, manually redo the high-cost, rated, and scale-sensitive. You set the tiers, not the AI, and the tier list doubles as the paper trail that defends your bid.
Worth knowing:Cost and risk drive the grade, not how clean the count looks. A big line item goes to manual-redo even when the AI seems sure, because that is the number that hurts if it is wrong.
Estimators moving from a checked takeoff to a real bidAnyone burned once by the single quantity that was wrongOwners who want a defensible number, not a fast guess
Start Lesson 7 →Find the gaps in the docs before they cost you
Catch change-order and RFI exposure early
The change order that hurts is the one found in the field
Caught before the bid, a missing detail or a spec-drawing mismatch is a quick RFI or a priced assumption. Caught after the pour, it is a change order and an argument about who pays. The AI cross-reads the documents and lists where they disagree or fall silent. Every flag is a lead you investigate against the source, and a clean scan is never proof the set is clean.
Worth knowing:A scan that finds little may mean the documents are clean or that the AI missed something dense. Absence of flags is not proof, so keep your own review on the high-risk scopes.
PMs surprised mid-job by a missing detailEstimators who keep eating spec-drawing mismatchesAnyone tired of late stacks of avoidable RFIs
Start Lesson 8 →Let the walk be the progress record
Document the site by just walking it
A fifteen-minute walk becomes a dated, tied record
Photo documentation wins disputes and protects payments, and it gets skipped because doing it well loses to the building. Walk-through capture records as you move and ties it to the plan, so the record builds itself. You confirm coverage after the walk and read the images yourself before certifying a payment, because an area you did not walk is a gap, and the camera shows only what it could see.
Worth knowing:Coverage equals the walk. An unwalked room is simply absent from the record, which is why a coverage check after every walk is the whole discipline.
Supers who never have time for a separate photo sessionPMs verifying sub work before a payment goes outAnyone who needed a pre-cover record they never made
Start Lesson 9 →Steps, hazards, controls, on the OSHA structure
A Job Hazard Analysis the AI helps you write
Beat the blank form, then make the JHA truly yours
A JHA breaks a task into steps, names each hazard, and sets a control. It is high-value and high-avoidance because the blank form is intimidating. The AI drafts the three columns on the OSHA structure, controls following the hierarchy. Then the work that matters: a qualified person reviews and revises it, the crew has input, and the signed document is your record, not the AI's draft.
Worth knowing:The stakes are higher than a toolbox talk, so the boundary is stronger: the AI is a drafting aid, not the safety authority, and a drafted JHA is not a completed one.
Foremen and safety leads staring at a blank JHA formSupers who put off the JHA until an audit forces itAnyone who wants audit-ready hazard analysis, faster
Start Lesson 10 →Cross-check one assembly before you build it
Catch a spec-vs-drawing conflict before the field
A five-minute catch at the desk replaces demolition
The most expensive place to find a spec-versus-drawing conflict is in the finished work. Give the AI the spec section and the drawing detail for one assembly you are about to build, and it flags where they disagree on dimension, material, rating, or method. You confirm each flag against the source, and a rated or structural mismatch is an RFI to the design team, never a field decision.
Worth knowing:The same conflict is a five-minute RFI before the build and a demolition-and-change-order after it. The catch is one cross-check before the crew starts.
Supers about to build a rated or structural assemblyLead trades who want the mismatch caught at the deskPMs tired of rework that the documents warned about
Start Lesson 11 →Judge the bigger claim as a skeptic
What's next: AI that re-plans your schedule
A real capability, and the most oversold corner of construction AI
Generative scheduling explores far more sequences than a human can and re-plans fast when work slips. It is also the most hyped claim in construction AI. Judge it on four tests: real constraints not a template, transparent and editable logic, true dynamic re-planning, and independent proof beyond vendor numbers. The schedule it makes is a proposal your experienced team pressure-tests, not an answer to obey.
Worth knowing:A mathematically optimal schedule can be operationally impossible. A sequence that ignores that two trades cannot share a floor is precise and wrong, and only experienced people catch that.
Owners weighing a generative scheduling pitchPMs who want an honest read on the hypeAnyone deciding whether a tool is worth a pilot
Start Lesson 12 →