McKinsey published its AEC artificial intelligence analysis on July 15 2026. The headline numbers are large. Roughly 39% of nonphysical work in construction sits within range of AI automation, and about 50% in architecture and engineering. More than 150 workflows across 25 AEC-related domains carry some automation potential. The value the firm puts on it is up to $228 billion a year in the United States by 2030, plus another $126 billion across Europe.
Those numbers are being quoted everywhere. The sentence that matters is not one of them.
"Capturing value from agentic AI requires more than adopting new tools; it means rewiring commercial and operating models."
That is McKinsey's own conclusion, and it divides the industry into what the report calls two camps: "those who leverage it to automate core tasks, and those who use it as a superficial productivity tool." Firms in the first camp, the report argues, are best positioned to reshape their businesses around AI. Firms in the second may hand that potential to partners, clients, or new competitors.
The gap the two camps are competing inside
The context explains the urgency. McKinsey's own productivity work puts construction's improvement at roughly 10% between 2000 and 2022, about 0.4% a year, against roughly 90% in manufacturing over the same period. Global construction demand is projected to grow from about $15 trillion in 2025 to $22 trillion by 2040.
So the industry is being asked to deliver half again as much work with a productivity record that has barely moved in two decades. That is the gap AI is walking into. It is also why the two camps diverge so sharply: a tool that makes an existing process marginally faster does nothing to that arithmetic, while a firm that changes how the work runs is operating on a different curve entirely.
The AGC's 2026 Construction Hiring and Business Outlook offers a rough read on where the industry currently sits. Of the 951 firms surveyed, 61% now use AI or plan to increase their investment, up from 44%. But 45% are deploying it for office and administrative functions, against 23% for estimating and 20% for design or preconstruction. Nearly twice as much adoption sits beside the core of the business as inside it.
AI value capture in construction is a commercial model question
Read McKinsey's description of what an agent actually does in the field, and the commercial dimension is right there in the sentence: agents update the project execution plan and send the field superintendent recommendations to adjust the work sequence to improve schedule, cost, and margin performance.
Margin. Not minutes saved.
The report's broader picture works the same way. In the agentic scenario it describes, a superintendent photographs an issue on a mobile device and within minutes agents compare that image against the current 3D model, the engineering drawings, the procurement records, and the schedule. What comes back is a recommendation. The decision stays with a person. And once routine work is handled that way, McKinsey's argument is that people move to decisions about pricing and risk, the areas that require judgment and that create differentiation.
Daniel Ahmoye, a partner in the practice, put the competitive consequence plainly: "Advantage is going to come from those who can redesign their work, their roles, how they operate, how they think about commercial models the fastest."
Notice what is absent from that sentence. No product. No platform. No model. The advantage McKinsey describes belongs to firms that redesign work, roles, operations, and commercial models. A tool purchase does none of those four things by itself.
"Buying an agent is a purchase. Rewiring the commercial model is a decision. Only one of them reaches your margin.
This is not a copilot story
McKinsey draws the line clearly. The next wave is not copilots that answer questions or draft documents. It is agentic AI, where autonomous software agents perform tasks, make recommendations, and coordinate work across entire projects with human oversight.
The report sequences that arrival in three horizons.
The first 18 months run through end-to-end workflow work: bid/no-bid analysis, estimating and pricing, proposal drafting, constructability and specification reviews, scheduling and procurement, RFIs and submittals, cost forecasting and change-order management, safety and quality, invoicing, document control, and compliance. Every item on that list is core contractor work, not back-office overhead.
Eighteen months to four years is where proprietary data becomes the differentiator. McKinsey points at the material a contractor already owns: RFIs, drawings, specifications, close-out reports, estimates, schedules, change orders, productivity history. Connected systematically, that becomes institutional advantage. Left scattered across systems, it stays a storage cost.
Beyond four years moves to the jobsite itself, with autonomous equipment and transportation coordination between factories, yards, and the site.
McKinsey also flags which skills change most by 2030: data entry, invoicing, and equipment inspection. Those are roles, held by people, on your projects.
McKinsey's playbook lands where governance does
The six-step playbook the report offers is worth reading against how most firms actually buy technology.
Step one is not a purchase. It is to prioritize three to five high-value workflows, rather than pursuing AI across the company. On partnerships, the guidance is to be aggressive but selective: share enough to create value while protecting the proprietary data, workflows, and know-how that differentiate the business. On build versus buy, McKinsey is blunt that AEC firms have historically struggled to build and scale software products, and that AI is advancing too quickly for most incumbents to rely mainly on internal development.
Then governance, framed as a scaling requirement: human oversight, audit logs, risk controls, data security, insurer-ready documentation, and role-based training. The reasoning is the part worth sitting with. Engineering decisions carry legal, contractual, and safety consequences, so AI recommends actions while experienced professionals remain responsible for approving designs and managing outcomes.
We have been making that argument to construction leaders for two years. It is useful to have McKinsey making it too.
Path 1 is the second camp. Not because those firms chose badly, but because administrative work is the easiest place to install something that asks nobody to renegotiate how the work runs. Core workflows resist that, since changing them means deciding who approves what, and that is a leadership conversation before it is a software conversation.
Which camp are you buying into
McKinsey names the destination. Here is the order that gets you there.
Questions first. Which problems does the firm actually need to solve? Which of those can AI meaningfully move? And which are broken processes that would only do more damage running faster? Pick your three to five from the answers, not from a vendor's feature list.
Governance second. Decide what AI handles alone, what it recommends for a person to approve, and what a person leads with AI supporting underneath. Every approval point gets a named owner. That is the same human oversight McKinsey's playbook requires before scaling, written down before you buy rather than after. We cover the mechanics in Govern AI Before You Buy It, and the underlying question in What Should Humans Still Own?
Then the Workflow Map™. The visual blueprint recording where your people decide and where AI executes. It is the artifact those first two steps produce, not the place the work starts.
McKinsey's talent note closes the loop: the future belongs to professionals who can design the collaboration between AI specialists and domain experts. In a mid-size general contractor, those domain experts are your estimators, superintendents, and project executives. They are the people who have to design this, which means they have to be in the room while it is decided.
The $228 billion is real. So is the 39%. Neither arrives with your name on it. What decides your share is whether you rewired the work, or bought something that ran alongside it. Our Discovery engagement is built around that sequence, and it starts with questions rather than a product.
Humanity for what matters. AI for everything else.
Sources
Daniel Ahmoye, Erik Sjodin, Jose Luis Blanco, with Aleem Mawji, Marita Winslade, and Patrick Rogers, "How agentic AI is transforming the AEC industry," McKinsey & Company Engineering, Construction & Building Materials Practice, July 15 2026 · McKinsey & Company, "Delivering on construction productivity is no longer optional" · Construction Dive, "How AI automation can fit into construction workflows: McKinsey," July 22 2026 · AGC of America and Sage, "2026 Construction Hiring and Business Outlook," January 8 2026, n=951