Vancouver's construction associations spent this fall asking a sharper AI question than "should we use it?"
In late September, the Vancouver Regional Construction Association ran an AI LAB session called Toolkit for Construction: Leading in the Age of AI. The first promise on the agenda was to help leaders "decide what to automate, assist, or keep human." The same session offered to help them ask smarter questions of their IT, HR, insurers and vendors, and to build a 90-day AI starter plan.
A few weeks earlier, the VRCA's Builders' Bench podcast asked a version of the same thing in its Construction AI Reality Check series: what tasks could disappear from a project manager's day, and what parts of construction will remain fundamentally human?
That's the right question for a Lower Mainland GC or trade contractor. Not whether AI is coming, but which jobs in your office and on your sites it should do on its own, which it should help a person do, and which it shouldn't touch.
Where the question came from, and the limits of the sources
An event agenda isn't a survey. It tells you what a regional association thinks its members need to hear, not how many contractors are asking or what they decided. The podcast is a conversation with a software company's AI specialist, and Procore has an interest in contractors adopting AI.
The same theme shows up elsewhere in the region, though. At BUILDEX Vancouver in February, a session titled AI tools you can use tomorrow was pitched at small and mid-sized contractors and built around three live examples: capturing the conversation on the drive home from an estimate visit, turning messy site notes into a structured plan, and building a simple assistant that knows your trade's standards and terms. At the same show, a vendor's vice president gave a talk on when construction firms should not use AI. That last source sells automation software, which makes its list of places to avoid AI more interesting, not less.
Nationally, a spring 2026 SiteNews and Procore survey found only 10% of Canadian construction respondents report widespread AI use across multiple workflows. Most firms are still sorting tasks one at a time. None of these sources tells you which of your tasks belongs in which bucket. That part is yours.
Why the sort matters in the Lower Mainland right now
BuildForce Canada's 2026 to 2035 outlook for British Columbia describes two markets moving in different directions. In the Lower Mainland, residential construction employment is projected to fall 15% from 2025 levels by 2035, and non-residential employment to slip 3% as tracked engineering projects wind down. Province-wide, non-residential investment is expected to peak in 2027, with ICI work like the Richmond Hospital expansion still elevated.
The people side is tighter. The same outlook says B.C. will need to recruit 48,200 construction workers over the decade, and 46,700 workers, or 22% of the current construction labour force, are expected to retire. Even with 40,800 new workers under 30, the province could be 7,400 short by 2035.
That mix, softer volume in some sectors and a steady loss of experienced people, is where the automate, assist or keep human question earns its keep. The region is also putting money into the skills side. The VRCA is working with Simon Fraser University's DIAL initiative on a construction substream of its Applied AI for Professionals program, with funded seats, aimed at helping people understand where AI adds value and how to fit tools into existing workflows. Training helps. It still leaves each firm to decide which of its own tasks go where.
Get it right and your senior people spend more of their week on the work only they can do. Get it wrong and you've handed a judgment call to a tool at the exact moment the person who'd catch the mistake is retiring.
What to automate
Automate means the tool does the task and a person checks the output on a schedule, not line by line. That's a narrow list, and it should be.
Good candidates share three traits. The task happens often, the inputs are consistent, and a mistake is cheap and easy to spot. In a GC's office that usually means things like:
- Filing and naming. Sorting incoming drawings, submittals and photos into the right project folders.
- First-pass summaries. Turning recorded coordination meetings into draft minutes and action lists.
- Status chasing. Flagging submittals or RFIs that have sat past their due date.
None of these makes a decision. Each one removes a chunk of repetitive work that someone is doing by hand today. If the tool gets one wrong, the cost is a misfiled document or a missed reminder, and the next check catches it.
Automate still needs an owner. Someone has to look at the output every week, notice when it drifts, and switch it off if it starts filing things in the wrong place. If nobody can be named for that job, the task isn't ready to automate. Move it to the assist list until someone can.
What AI should assist with
Assist is the biggest bucket for most contractors. The tool drafts, a person decides, and nothing goes out the door without a name on it.
This is where the BUILDEX examples sit. A recorded conversation on the drive back from a site visit becomes a scope outline the estimator corrects. Messy notes from a walk become a draft deficiency list the superintendent edits. A trade-specific assistant drafts the first version of an RFI or a delay notice, and the PM rewrites the parts that matter.
The BC Construction Association's free AI Hub for contractors puts the rule plainly: review anything AI creates before you use or send it. It also names two things to never paste into any AI tool, free or paid. One is your cost breakdown: labour rates, margins and unit prices. The other is documents that aren't yours to share, because many owner and GC contracts carry confidentiality clauses.
The useful question isn't "can AI do this?" It's "who signs it when AI has helped?"
Assist work is where most of the time savings live, and it's also where the habits get set. If your people learn to treat every draft as a first draft, the tool gets safer as it gets used more. More on that habit in Don't Believe Everything the Robot Says.
What stays human
Some work shouldn't be handed to AI at all, and it's worth writing that list down before anyone buys anything.
The BUILDEX talk on when not to use AI named high-stakes, nuanced decisions: negotiations and site relationships where the outcome depends on trust. For a contractor, that list usually includes:
- Price and risk. What you bid, what contingency you carry, and which terms you accept.
- Safety calls. Whether a task is safe to start, and what changes when conditions do.
- People. Hiring, discipline, and the conversation with a sub who's falling behind.
- The client relationship. The phone call when something has gone wrong.
AI can prepare the background for any of these, such as a summary of the subcontract or a draft agenda for a tough meeting. The decision and the conversation stay with a person who's accountable for them. The broader version of this question, what software is allowed to do without asking, is covered in Your Software Can Now Act Without You.
Three signs a task isn't ready for AI yet
The same BUILDEX speaker gave three warning signs, and they hold up for most contractors.
The data is a mess. If your project logs, submittals or safety documents live in five places with no consistent naming, AI won't clean that up. It will just produce confident output from bad inputs. Fix the folder structure first.
The time saved is tiny. A task that takes someone five minutes a month isn't worth automating. Look for the jobs that eat hours every week, not the ones that are merely annoying.
Nobody can say what "good" looks like. If two of your PMs would write the same RFI two different ways and both think theirs is right, a tool will just add a third version. Agree on the standard before you ask a tool to meet it.
None of these means stopping. They mean the task belongs in a later round, after the groundwork is done. The trade-business version of this sort, by department, is in Where AI Actually Fits in a Trades Business.
What to do with this
Take one week of work from one project and list the recurring tasks your PM, coordinator and superintendent do. Don't filter yet, and include the small jobs that nobody thinks of as work, like renaming files or chasing a missing submittal.
Then mark each one A, B or C:
- A, automate: frequent, consistent, cheap to get wrong.
- B, assist: AI drafts, a named person reviews and sends.
- C, keep human: price, risk, safety, people and client decisions.
Pick two from the A list and two from the B list. Time them before anything changes, write down who reviews the output, and check again in 60 to 90 days. Anything on the C list gets written into your company's AI rules so nobody automates it by accident.
If the BCCA AI Hub's Should I use AI? check helps your team talk about it, use it. The VRCA's AI LAB format, a 90-day plan built around what to automate, assist or keep human, is a sensible shape for that first round.
Working through that sort for one business, with the people who actually do the work, and deciding what governance needs to be in place before anything changes, is what the AI Discovery and Readiness Assessment is for. It's a paid engagement with a defined scope: a discovery call, structured input from the people closest to the work, and a written report on where AI is relevant to your operation, what to do first, and what needs to be in place before you start.
