What AI changes on a job site by 2027, and what it doesn’t

A look at which parts of running a job will change first, and which will stay in a superintendent’s hands.

Cover image for “What AI changes on a job site by 2027, and what it doesn’t”

·

What’s coming

What’s coming

·

Construction

Construction

Most predictions about AI in construction start with robots. The first real change is quieter. It’s in the hours around the building: the RFIs and change orders, and the calls to chase an answer.

Over the next year, we expect that follow-up to move off people’s plates. Here is our reasoning, and where the limits are.

Why paperwork goes first

The AI that has improved fastest is the kind that reads and writes language. A model trained on a very large amount of text can now read a foreman’s message or an architect’s reply and understand what each is asking for. A robot that can frame a wall in the rain on an uneven slab is a much harder problem, and a different field of research.

Construction happens to run on language. A mid-size job produces thousands of messages, and most of the coordination lives in them. So the technology that matured first lines up with the part of the job that takes the most office hours.

The pieces are already in use. An RFI gets drafted from a text message. A change order is written when the architect answers and priced against the estimate. Time cards photographed in the field reach payroll without anyone typing them.

What we expect by 2027

We expect drafting to be ordinary. A project manager at a ten million dollar contractor will not write a routine RFI from scratch, any more than they would draw a schedule by hand.

We expect the schedule and the budget to be current by default. Today they are right on the day someone updates them. When updates are made from the messages as they arrive, the status of a job is something you look up on a Tuesday afternoon, and the Monday meeting stops being the only time it is known.

We expect the office to carry more jobs. Today each new job brings its own load of handoffs. When software carries those, the same office supports more of them, and the experienced people in it spend their time on review and on clients.

And we expect owners to ask for it. An owner who has worked with one contractor whose change orders arrive priced and documented within a day will expect that from the next one.

What stays in a superintendent’s hands

Judgment stays where it is. Someone still walks the site and decides whether the work is right. Someone still keeps the relationship with the owner, and with the inspector who wants a call first.

None of that is going to a machine, and it shouldn’t. A model can read every message on a job and still know nothing about how a wall looks when it is out of plumb.

Sequencing stays too. Deciding which trade goes first when two are behind is a negotiation between people who will work together again, and it rests on trust built over years of jobs.

A year is a short time. We’re confident in it because every piece is already running on real jobs, and we’ll keep writing about what we see as it happens.

What should we cover next?

Tell us what you’re trying to figure out about AI in your work. The good questions become our next pieces.

What should we cover next?

Tell us what you’re trying to figure out about AI in your work. The good questions become our next pieces.

What should we cover next?

Tell us what you’re trying to figure out about AI in your work. The good questions become our next pieces.