Every construction software vendor now has AI on its homepage, this one included. Most of it is a chat box. Some of it removes real work each week. Telling the two apart before you buy is worth doing, so this guide is organised around what is shipping today rather than what is demonstrated at trade shows.
We build one of these products, so treat our enthusiasm accordingly. The second half of this article is what our own AI can read, what it cannot, where a person has to approve, and one estimate worked through to the cent, so you can check our claims the same way we suggest checking anyone's.
Where AI genuinely helps a builder today
Three uses have moved from demo to daily work for residential builders and contractors.
- Drafting estimates. A project brief becomes itemized labour, material and subcontract lines in seconds. The builder still checks quantities, prices and markup, but the blank page is gone.
- Answering questions about your own jobs. Which invoices are outstanding, which jobs are behind, who is free on Thursday, in plain language instead of a report.
- Drafting correspondence and reading paperwork. Change order explanations, proposal copy and follow-ups, and pulling the vendor, amount and dates off a supplier's bill.
Where it is still mostly a demo
These are the claims to press hardest on, because the gap between the video and the shipped product is widest here.
- Blueprint and takeoff reading. Extracting accurate quantities from a drawing set is hard, and accuracy varies with drawing quality. Ask to test it on your own plans, not the vendor's sample.
- AI scheduling. Generating a plausible-looking Gantt chart is easy. Generating one that respects trade sequencing, lead times and crew availability on your jobs is not.
- Cost forecasting. Predicting a job's final cost needs years of your own clean historical data. A vendor cannot give you that on day one, whatever the model.
- Autonomous agents. Software that takes actions on your jobs without review is being marketed well ahead of where most builders should be comfortable.
One estimate, worked through
The brief: "Ensuite renovation: retile shower and floor, new vanity, toilet and fixtures, repaint." That goes into Bridge AI's estimate builder with the project type, the size, a finish level and the job's province, here Alberta. The model returns a draft of itemized lines. That draft is not reproducible: the same brief can come back with different lines on a different day, which is exactly why it is a draft.
What follows the review is reproducible. After the builder corrected the draft (52 sq ft of tile measured, not the 45 drafted, plus waste; the fixture allowance set at $2,800), the 8 lines below cost $9,407.00. AI drafts are created at 15% markup; the builder changed it to 18%. From there the app, not the model, works out the price and the tax, and it returns the same figures every time for the same lines. They are the same figures as on our estimating page and in the free estimate template.
| Item | Amount | Worked out by |
|---|---|---|
| Cost of the 8 reviewed lines | $9,407.00 | The builder, after editing the draft |
| Price at 18% markup on cost | $11,100.26 | The app's pricing code |
| GST at 5% (Alberta) | $555.01 | The app's tax code, from the job's province |
| Estimate total | $11,655.27 | The app |
What Bridge AI can read
An assistant is only as good as what it is given, so this is the list, from the code. Each feature reads only what the signed-in person could already see in the app.
- The estimate builder reads the project type, size, finish level, your scope notes, the job's province or state and your business name, and optionally one attached plan, drawing or photo (PDF or image, up to 12 MB).
- The chat reads a snapshot of your workspace: up to 12 projects with status, progress, client and due date; up to 10 open leads; up to 10 open invoices; your top 6 clients by amount invoiced; the active crew; and the next 7 days of the schedule, with double-bookings, unassigned work and hours scheduled against hours logged.
- Money figures in the chat (contract values, balances, pipeline) are included only for roles allowed to see financials. A crew member's answers leave them out, as their screens do.
- The chat does not read documents, daily logs, photos, messages or estimate lines. Ask it about those and it should say it does not have the data.
- The bill scanner reads one photo or PDF of a supplier's bill (up to 8 MB) and returns the vendor, invoice number, amount and dates.
How an answer points to the record
The chat is told to answer questions about your business only from the snapshot, to name the record a fact came from by its code or number, and to say "I don't have enough project data to answer that" rather than guess. Under each reply, the app links every project and invoice the answer mentions, so you can open the record and check it.
When the model is unavailable, the chat says so. For a question about your own receivables, jobs or job costs it then shows what the live records say, worked by the app's own calculations, with links to those records, and nothing else. Every chat reply, estimate draft and piece of proposal copy has a report button, so a wrong answer can be flagged with the text it came from.
Where a person decides
Nothing the AI produces becomes a record, or reaches a client, without a person.
- An estimate draft is only a proposal on screen. It is saved when you press "Looks right, create this estimate", and then as a draft estimate marked AI-generated.
- Tax is never the model's: the app calculates it from the job's province or state, or your workspace's tax override, when the estimate is created and priced.
- Sending is a separate step you take. The client approves in the portal, and approval sets the contract value.
- A scanned bill is shown for review and saved only when you confirm it, because an amount read off a photo is a suggestion.
- Proposal copy is drafted only from the facts already on the job, and you edit it before it goes anywhere.
- The chat replies with text. It cannot create, change or send anything, and it cannot move the schedule.
If the model is down, or the size looks wrong
When the model is unavailable, the estimate builder does not invent prices. It returns a scope checklist for the project type, with quantities from the size you entered and every unit price blank for you to fill in, and it says that is what happened. If the size you entered is well outside the usual range for the project type, say a 1,000 sq ft bathroom, the draft carries a warning, because a typo in the size scales every quantity with it.
An attached plan is not a takeoff
When you attach a plan, drawing or photo, the model reads it as context for quantities and scope. Nothing measures the drawing, and nothing checks the quantities against it, so treat any quantity from it as one more line to verify. A takeoff tool that measures plans on screen is on our roadmap, along with AI scheduling and AI cost forecasting. None of the three is available today, and they are marked Planned on our comparison page. If you need them now, other platforms in the category are further ahead.
Watch the recording
There is an unedited screen recording of the estimate builder on our homepage, published 21 July 2026: a different bathroom brief turned into 9 lines, with GST at 5% and the 15% starting markup, for $48,949 in total. Its lines are not the ones above, which is the point: the model's draft changes with the brief and between runs, and the review is where it becomes your number.
Five questions to ask before you believe it
These separate shipped capability from a roadmap slide faster than any feature list.
- Is this available on my plan today, or is it in beta or planned? Ask for the answer in writing.
- Can I test it on my own drawings, my own estimates and my own data during the trial?
- What exactly does it read, and can it see anything my own login cannot?
- Does the output stay a draft until a person approves it, particularly anything a client sees?
- When it states a fact about my business, can I open the record it came from?
The rule that matters: AI drafts, the builder decides
An AI-generated estimate is a starting point, not a binding quote. That is not legal boilerplate. It is the difference between software that saves you an hour and software that sends a client a number you did not check.
Any tool that puts AI output in front of a customer without a review step is solving the wrong problem. The hour spent writing the first draft is worth removing. The minutes spent checking it are not.