AI in construction won’t arrive as one big event. It will show up as a thousand small changes in how paperwork moves, who writes the first draft, and who checks it. Nobody can tell you what 2036 looks like. I can tell you what the measured evidence says today, where it’s thin, and what I’d do with it.

One disclosure. I build custom AI tools and teach AI to contractors, so weigh my opinions with that in mind. The numbers below come from independent sources, and I label my opinions as opinions.

The starting point: slow productivity and too few people

Productivity. The McKinsey Global Institute found that construction labor productivity grew about 1 percent a year over the prior 20 years. The total world economy grew 2.8 percent a year, and manufacturing grew 3.6 percent. It estimated that matching the broader economy would add about $1.6 trillion in value. That report was published in 2017 and covers the world, so treat it as direction, not a current measurement.

Labor. Associated Builders and Contractors (ABC) says the industry needs to attract 349,000 net new workers in 2026, down from a 2025 projection of 439,000. It says most of the 2026 demand comes from retirements, not growth.

The AGC and NCCER 2026 Workforce Survey (1,830 respondents, July and August 2026) shows the pressure on the ground:

87% of firms had openings for hourly craft workers.
Of those, 88% said the jobs were as hard or harder to fill than a year earlier.

My read (opinion): the question for the next decade isn’t who gets replaced. It’s how you do more with the people you can find, and how you keep what your retiring people know.

Where contractors are today

Two AGC surveys show the shift. Both are self-reported.

Survey	Finding
AGC 2026 Outlook (951 firms, Nov. 4 to Dec. 15, 2025)	61% use AI or plan to increase investment, up from 44% a year earlier
Same survey, current uses	Office and admin 45%, estimating 23%, design or preconstruction 20%, recruitment, training, or HR 16%
AGC and NCCER 2026 Workforce Survey, expected uses	Estimating and bidding 67%, document and contract review 66%, scheduling and planning 62%, project management 49%
Same survey, five-year outlook	51% expect a positive impact, 21% mixed, 8% negative

The pattern is clear. Contractors expect AI to help most with paperwork and bids. That matches where the hours go.

Custom AI solutions: where the value is likely to come from

McKinsey’s 2026 State of AI survey (1,719 respondents in 97 countries, May 4 to June 8, 2026) points to a gap between using AI and profiting from it:

Nearly nine in ten respondents use AI regularly in at least one business function.
37% attribute a positive EBIT impact to AI at the enterprise level.
Only 6% qualify as high performers.
Nearly three-quarters of high performers fundamentally redesigned workflows because of AI.

That’s cross-industry data, not construction. But it supports a point I see as common sense. Buying a subscription doesn’t change a business. Changing how a task gets done does. A custom tool is a redesigned task.

A second study explains why a narrow tool beats a wide-open chat window. Harvard Business School and BCG studied 758 consultants. On tasks inside what AI does well, consultants using it completed 12.2% more tasks and worked 25.1% faster. On a task outside that frontier, they were 19 percentage points less likely to produce a correct answer. They still worked fast, and their answers still sounded convincing.

Translate that to a jobsite. A generic chat window will answer any question, including the ones it handles badly, and sound sure about it. My opinion: a custom tool keeps the work inside the frontier. It uses your documents and your rules for one task, and a person checks the result.

One more finding, from outside construction. An NBER study of 5,179 customer support agents found AI raised issues resolved per hour by 14% on average and by 34% for novice and low-skilled workers. It found minimal impact for experienced workers. My read: tools that carry your best practice help newer people most. As experienced estimators and PMs retire, a tool that holds your company’s rules could close part of that gap. The evidence comes from a different industry, so treat that as a hypothesis.

My own result: a four-day takeoff came back in about seven minutes, 97% accurate against the estimator’s count, on one real electrical drawing set. That’s one project and my own number. It holds when the business rules are documented and the drawings are clean.

Employee training: the cost nobody budgets

BCG’s 2025 AI at Work survey covered more than 10,600 leaders, managers, and frontline white-collar employees in 11 countries and regions. It found:

Regular use among frontline employees stalled at 51%, while more than three-quarters of leaders and managers used generative AI several times a week.
Regular use was sharply higher for employees who got at least five hours of training and had access to in-person coaching.
Only one-third of employees said they’d been properly trained.

This isn’t construction data, and these are office workers. But the pattern fits a question I hear from associations: how do we get the team to use it instead of nodding through a training?

My opinion: training is the constraint on AI in construction, not tools. Budget for hours per person, train by role, and practice on your own documents. Repeat it.

Safety: use AI for paperwork, not judgment

The Bureau of Labor Statistics recorded 1,032 fatalities among construction and extraction workers in 2024. Fatal falls, slips, and trips in that group fell 7.5% to 370 from 400 in 2023.

A 2025 systematic review in the journal Buildings (Badhan and Samsami) examined 122 peer-reviewed studies from 2016 to 2025. It found:

Computer vision models are the most common tools for personal protective equipment recognition and proximity monitoring.
A language model reached 93.7% accuracy on safety narratives from OSHA injury reports.
The barriers are real. Construction data is often incomplete and noisy. Tools can misalign with site protocols. Safety managers may distrust “black box” output. Many firms lack in-house expertise.

My opinion: the near-term win is documentation. Site-specific safety plans, activity hazard analyses, toolbox talks, and incident write-ups. A person still owns the decision. I teach a five-step check, G.U.A.R.D. (Gather, Understand, Assess, Risk-rank, Document), and every AI-generated safety document runs through it before anyone signs.

ROI: measure it or don’t spend it

I didn’t find an independent, construction-specific ROI study I’d trust, so I won’t invent a number. McKinsey’s 37% means most companies can’t yet attribute a profit impact to AI. So measure your own return on AI in construction:

Pick one task.
Time it on real documents, with and without the tool, including the human review.
Track errors caught and errors that got through.
Compare the yearly savings to the yearly cost of the tool and the training.

The formula: hours saved per task, times tasks per year, times your loaded hourly cost, minus the yearly cost of the tool and training. If you can’t fill in the first number from your own timing, don’t buy yet.

What I expect over the next decade (opinion)

Drafting moves to AI. Review stays with people. The job shifts from typing to checking.
Company knowledge becomes an asset. Your pricing rules and standards live in tools you own.
Training and policy separate the winners. Tool choice matters less than whether your team knows how to use what you have.

What to do in the next 12 months

Pick two repeatable, document-heavy tasks and time them today.
Write a short AI policy that says what data never goes into a public tool.
Train your people by role, at least five hours each, on your own documents.
Build or buy one narrow tool. Test it against your current process.
Review the numbers every quarter.

What this evidence can’t tell you

Most surveys are self-reported. Several studies come from other industries. The productivity data is from 2017 and covers the world. I didn’t use vendor research, and I found little independent construction-specific evidence on AI return. Plan for a decade of change, and expect the details to move.

FAQ

Will AI replace estimators and project managers?
Nothing in the sources above shows that. AGC’s contractors expect AI to automate repetitive tasks, raise productivity, and improve safety, while firms struggle to fill open positions.

What should a contractor automate first?
Start with repetitive, document-heavy work. Contractors in the AGC and NCCER survey expect the biggest use in estimating and bidding and in document and contract review.

How do I measure AI return on investment?
Time one task on real documents with and without the tool. Compare the yearly hours saved, at your loaded labor cost, against the yearly cost of the tool and training.

Is AI safe to use for safety documents?
Use it to draft. A qualified person reviews and signs. The research shows real promise and real barriers.

Sources

Want this applied to your company? Book a free consult with Iron Thread.

What AI Means for Large-Scale Construction Over the Next Decade