AI in Construction: Stop Asking Which Platform to Buy

The buying question construction operators have been trained to ask — which platform, which suite, which module — is the wrong one for the next twelve months. Two better questions have taken its place, and the firms that answer them first are going to look very different by 2028 than the ones that don't.

Here's what changed. The cost of what used to justify a construction software subscription has collapsed. Takeoff quantities, schedule generation, RFI triage, submittal tracking, spec-compliance checks — the computational work vendors priced into per-seat licensing can now be handled by a competent mid-market team with a frontier model and a couple of weekends of setup. That is not a forecast. It is already happening on projects you would recognize. The computation was never the part of the business that made your firm hard to replace. It was just the part your vendors were expensive enough to charge for. That pricing model is coming apart.

What isn't coming apart is the senior PM who reads a bid package, taps a line in Division 15, and says the mechanical looks light. Nine times out of ten she has called it correctly. Ask her which specific number is off and she'll walk over to the file cabinet. That judgment is assembled out of every job she has watched go sideways — the change order that ate the fee, the equipment vendor whose lead time nobody believed until the crane arrived on the wrong Tuesday, the spec-section interpretation that cost the firm three jobs in a row before anyone thought to write it down. The read lands before the explanation catches up. And that read is the asset — the one no software has ever priced correctly, and the one every AI conversation ought to start with.

Meanwhile the AI you keep hearing about is not going to help you if your operational history lives across three systems that don't talk to each other, a shared drive from 2019, and a handful of long-tenured heads. Per the AGC's 2026 Construction Hiring and Business Outlook, 45% of firms have deployed AI for office and administrative work, but only 23% use it in estimating — the part of the business where wrong answers actually cost money. The gap between those two numbers is the story. The firms that will close it are the ones consolidating their systems, standardizing project data, and forcing their operational history into one place a model can read. Everyone else is paying a monthly fee to arrive at wrong answers on a shorter timeline.

So the two questions.

First: which decisions on your team are being made by pattern recognition that no system currently records — and how do you build the practice of capturing that expertise into the way your firm operates, so the technology you eventually deploy adapts to your methodology instead of the other way around? The estimator's read on when a bid feels off. The superintendent's judgment on which crews handle a difficult pour. The PM's mental list of subs who always come through in the last two weeks of a job. Document those as structured patterns — the read, three examples, the exceptions, the reasoning behind the call. Fold that documentation into how newer people are trained, so the next generation isn't rebuilding institutional judgment from scratch. Then upskill the whole team to work alongside AI tooling as it matures, rather than around it. This is not about capturing knowledge before people leave. It is about making sure the way your firm actually operates — the methodology, the judgment, the trained instinct — is what shapes the tools you adopt, not the reverse.

Second: which of your existing vendors can expose their data through an open interface that whatever AI you choose can query directly? Not "does it export to Excel." Direct, machine-readable, live. If a vendor cannot give you a clear roadmap on that in the next twelve months, factor them into your two-year technology planning as a risk to actively hedge — not as a partner to write off, but as a line item that needs a proactive lookahead. Build a vendor review into your annual planning cycle: forecast which of your current tools will still be in the stack in three years, which will need replacing, and which will need to run alongside more open alternatives. The consolidation coming for construction tech will reward the firms that saw it in their forecasting long before it arrived, and it will punish the ones that treat vendor selection as a one-time procurement decision.

Most of the noise in construction tech right now is about speed — faster takeoffs, faster proposals, faster schedule updates. Speed is fine. It is not a differentiator. What separates the firms that thrive over the next decade from the ones that get consolidated is whether the operational judgment their business was built on has been made durable — captured in the way people are trained, in the systems they work in, and in the data those systems expose to the next generation of tools.