A project manager loses a Thursday to a budget that already exists. The real number is out there—in the superintendent’s head, in a change order nobody logged, in two email chains and a spreadsheet three versions behind. She spends the afternoon assembling it, and by the time the owner has a figure to trust, it’s already moved.
That afternoon is the story—not the software she used, but the hours of a smart, trained person turning scattered information into a clean answer. And the most prestigious companies in the world just told you what happens to it.
McKinsey is cutting 3,000 to 4,000 jobs in 2026—around 10% of its people, its deepest cut since 2008. Bain, BCG and Deloitte are running quieter versions of the same play. The firms describe it, carefully, as a response to AI productivity gains.
Here’s the takeaway for construction: This is not a technology story. It’s an operating-model story. Whether you run a GC, a specialty trade or a multi-site build program, the same compression is coming for the same layer of your business—and the firms that redesign for it first will take share from the ones that wait.
What Work Actually Disappeared at McKinsey?
McKinsey didn’t lose its partners. It lost the layer beneath them — the analysts who spent weeks pulling research together, building models and turning it into the deliverables clients paid premium rates for.
That work compressed—not because the people were weak, but because gathering, synthesizing, and formatting information is exactly what AI now does in minutes.
Now walk your own business. Who builds the first-pass estimate on a new pursuit? Who reconciles budget-to-actual across a dozen subs and a stack of change orders? Who writes the closeout package the owner uses to release retainage?
That isn’t McKinsey analyst work, but the shape is identical: collect, reconcile, synthesize, report. Skilled work, done by capable people—and exactly what AI is compressing faster than most builders planned for.
Where Does Construction Carry the Same Exposure?
This isn’t a forecast. It’s already on your jobsites. In ServiceTitan’s 2026 survey of 1,000-plus construction leaders, 38% now report measurable impact from AI—more than double the 17% a year earlier. Where is it landing first? Cost estimating, budgeting, and bid management. The back office.
Estimating that used to take three days turns around in an afternoon. The owner update built at 9 p.m. the night before the OAC meeting now assembles itself from the project platform in real time—without the 9 p.m.
Why Won’t Buying an AI Tool Fix It?
Here’s where most firms stall. They see the compression, buy a platform, run it on one project, and call it a strategy. The tool works—and almost nothing changes, because it’s bolted onto an operating model nobody redesigned to use it.
The technology isn’t the problem—the order is. Layer an AI strategy onto a fragmented operating model and it won’t create leverage. It just speeds up the mess.
Faster reporting on a job where nobody’s sure who can green-light a change order just documents the schedule blowing up faster. Cleaner data still routed through an owner who approves everything is a bottleneck with better lighting. Neither touches the debrief that never happened, so the same field miss loads into the next job.
McKinsey ran this experiment at scale. What survived the automation was judgment—the partner who could tell a CEO something the CEO hadn’t already worked out. The scaffolding beneath that judgment was the exposure, not the asset. Your company is built the same way. The only question is whether you can name where your own exposure lives.
Does AI Replace People–Or Your Need for Judgment?
Treat AI as a headcount play and it will cost you—especially now. The AGC puts the industry short about 500,000 workers in 2026, with roughly 41% of the craft workforce set to retire by 2031. You can’t hire your way out of that. The firms that win will get more from the people they already have.
AI doesn’t take the 6 a.m. call when the steel showed up wrong and the crane is booked three hours. It doesn’t decide whether to pull a second crew onto the deck when the framing sub is short four and the inspection is Friday.
And it doesn’t carry the memory of two hundred jobs—the instinct that knows what a doomed schedule feels like in week three, before it’s a crisis in week 10.
What AI does is hand that judgment more room to operate. That line—between the work AI compresses and the judgment it amplifies—also is your sharpest lens for the next hire.
The person worth hiring now thinks clearly on partial information, holds up under pressure, and runs alongside AI tools without a manager hovering. Keep writing 2019 job descriptions and you’ll staff a 2019 company in a market that already moved.
Why the Order Matters: Design First, Then Automate
In construction, the AI discussion is stuck at the tool layer: which platform, which estimating software, which reality-capture demo looked sharpest at the last trade show. That discussion isn’t useless. It just starts in the middle.
Take a normal day. A PM running six jobs gets a budget-update request on Job A. The number means chasing the super, the drywall sub, and an electrical change order revised twice and buried in someone’s inbox—two hours and it’s stale on arrival.
Meanwhile a real call on Job C—a second-tier sub, because the primary is short-crewed—has sat two days: she wasn’t sure she could approve the cost, and the owner’s traveling.
None of that is a technology problem. It’s a decision-rights problem in a technology costume. Who owns that budget update? What’s supposed to happen the moment a sub comes up short? Does approval authority sit where the org chart claims? Until those have clean answers, AI just hands you tidier versions of the same confusion.
Fix the model first and the same day runs differently. Everyone knows who decides what; information lands with the right person because the system was built to route it there.
Now AI earns its keep—it shrinks the gap between knowing something and acting on it. The budget update goes from a lost afternoon to 15 minutes. The crew shortfall surfaces before pour day. The change that once appeared in week ten shows up in week three, while there’s still margin to defend.
Whatever operating model AI lands on, it accelerates. Land it on a good one and the gains stack. Land it on a broken one and you’ve bought a richer data feed on a problem you still can’t close fast enough to matter.
Three Questions Worth More Than Any Software Demo
Construction runs on thin margins, and they’re getting thinner. The average general contractor nets 5% to 6%; only the top performers clear 10% to 12%. Materials squeeze from one side—steel is up roughly 20% year over year, aluminum more than 30% and 53% of contractors name materials costs a top concern for 2026.
Labor squeezes from the other. When a job turns sour in its last three weeks, it eats the margin the whole thing was priced to earn. Most firms absorb the hit, close out and move on—cause unfixed, next job already underway.
That’s the exposure AI will either close or widen. No software demo needed to work through these three:
- Which margin is bleeding out through calls made too late? Not the overrun you already caught, but the one sealed in by week three because nobody was sure they were cleared to name the constraint. AI can raise that flag in week one—if someone has the authority to act on it.
- How many hours are your best PMs and estimators pouring into work AI will absorb inside two years—and what does that push out? The senior PM hunting a sub’s invoice or rebuilding a number that already lives in three inboxes, isn’t tending the owner relationship that hands you the next award. Name it for what it is: not a productivity leak, a margin leak.
- How often does a call still land on the owner’s desk when it should have been settled two rungs down? Tally what that costs—in speed, in margin, and in whether the company can function when you’re not in the room.
Answer those three honestly and you’ll understand your AI readiness better than any technology audit will—and see precisely which parts of the operating model must change before AI spend pays back.
McKinsey couldn’t see the exposure in its own model until the cuts were on the table. You get to study it on someone else’s balance sheet first—with enough lead time to build around it instead of reacting once it’s yours.
That lead time has a shelf life.
Jane Gentry founded Jane Gentry & Company, a strategic operating-model advisory firm working with mid-market CEOs and senior leadership teams in construction, manufacturing, events, and professional services. Her work centers on three things: how the work actually gets designed, where decisions really get made, and how to turn AI into leverage rather than faster drag. If any of this is already live in your business, start with a 60-minute diagnostic conversation. Reach her directly at jane@janegentry.com.
