On 15 July, McKinsey published eleven pages on AI in architecture, engineering and construction, and put the reassurance and the threat in one sentence: AI is "unlikely to be an extinction event for AEC firms, but it could meaningfully change who leads the industry." Eleven days on, the Nordic trade press has yet to print a line about it. The vendor decks are less patient, and the board agendas fill in the autumn. This is the report read from the buyer's chair.
The publication
McKinsey opens the report with a scene it is careful to label. A field superintendent discovers that prefabricated pipe spools no longer fit after a late engineering change. Today, the text says, that might trigger days of RFIs, drawing reviews, procurement checks and schedule updates while crews wait. In the agentic future it sketches, the superintendent photographs the problem, AI agents check the image against the model, the drawings, the procurement records and the schedule, and the project team sees costed options within hours. "Human experts remain in control," the text notes. It is an example, "among many potential transformations", and it is honestly framed: no such pipe spools exist. The workflow they stand for exists at every firm that will read the report.
The byline is a collaborative effort by six authors across five offices, and the Stockholm name is worth registering: Erik Sjödin, a partner and one of the leaders of McKinsey's engineering and construction work, thanked in the acknowledgments of the firm's 2024 productivity article, credited as an author on this one. International trade press picked the report up within a week. In the Nordic press, as this is written, it has landed in silence. The silence is useful. A firm that reads the original before the sales decks arrive gets to form its own reading, and this autumn that is a real advantage, because the decks will quote the report accurately and select from it freely.
The history
The report is the third act of a documented sequence. In February 2017 the McKinsey Global Institute measured the gap in Reinventing Construction: productivity growth of 1 percent a year over two decades against 3.6 in manufacturing, one of the least digitised sectors in MGI's own index, an estimated 1.6 trillion dollars of value added on the table. In August 2024, "Delivering on construction productivity is no longer optional" declared itself a continuation and moved the stakes to a shortfall: up to 40 trillion dollars of cumulative undersupply by 2040. The 2026 report footnotes the 2024 article and does what neither predecessor did: it sorts. Early gains "will likely soon be table stakes"; firms that treat AI as a superficial productivity tool "could cede its potential to partners, clients, or new competitors"; the closing line has advantage drifting to "competitors, vendors, AI-native startups, or even clients themselves." The sorting is of companies, and the same text expects those companies to become more dependent on expert judgment as the routine layers automate. The report is not a forecast. It is a memo to the board.
The numbers
The memo's numbers will travel, and a buyer should sort them before the vendors do. The billions are conditional: the industry "could unlock roughly $228 billion by 2030 in annual value in the United States", the impact on Europe's construction sector alone "could equate to roughly $126 billion", and AI has "the potential to automate" 50 percent of nonphysical work in architecture and engineering and 39 percent in construction. Could, potential, roughly: the verbs are part of the finding, and they will appear on autumn slides with the verbs removed. The series that holds is drier. Construction productivity improved 10 percent in total between 2000 and 2022 while manufacturing improved 90; global construction output ran about 15 trillion dollars in 2025 with projected demand heading for 22 trillion by 2040; the cumulative shortfall on today's trajectory runs up to 40 trillion. And the most informative part of the report is the least quotable. For the first eighteen months, its highest-value workflows are bid and no-bid analysis, estimating, specification review, code and standards interpretation, schedule development, contract-obligation tracking, RFI and submittal triage, invoice validation. The robots keep the conference keynotes; large-scale site robotics sits "potentially a decade away" in the report's own timeline. The near term is document work, and the report's advice for it is structural: connect the work across teams and stages instead of optimising isolated tasks, because "the firms that succeed with AI in the near term will be the best at orchestrating agentic work across workflows."
The translation
Translate the list and it stops being a technology forecast. Bid and no-bid analysis is Friday's call on the frame agreement. Specification review is the granskning round. Contract-obligation tracking is the ÄTA file and the AB 04 notice deadlines. Estimating and scope normalisation live in the calculation department, and RFI triage is the fråga-svar log of any large delivery. McKinsey's near term, in other words, describes the sector's existing week, with the production of the documents moved to machines and the judgment left where it was.
The list is also further along than the report's careful tense suggests, and here we can add what we see. At firms we work with, the first time a whole deliverable is handed over, a noise study, a control plan, a tender review, it tends to come back the same day. The revealing part is what happens next: the firm's existing review discipline absorbs it. The granskning that has always stood between a junior engineer's draft and a signature stands just as well between a machine's draft and a signature. The engineer reviews, corrects and signs, exactly as the quality system always required. The profession spent a century writing down its methods and its controls, and the new producer arrives to find the reception already built.
That experience reframes the window. Eighteen months measures the sector's adjustment time rather than the technology's arrival. The trap in the near-term list is to meet it with a chat assistant on every desk and call the box ticked: the report is explicit that treating AI "as a more superficial productivity tool" is how a firm ends up ceding, and its cure is structural, redesign a domain, hand a deliverable over whole, keep the judgment. Which raises the question the report answers best: what exactly to keep.
The layer
The report's centre of gravity is one sentence, worth quoting whole.
As agents become embedded across functions, more value and profit will go to firms that control three things: proprietary project data, the workflows where decisions get made, and the ability to charge for outcomes rather than hours.
In plainer terms: your archive, your decisions, and your invoice. On the archive, the moat forms while the work happens. Project files record what was decided; information captured during the work records how and why, the alternatives weighed, the assumptions made, the trade-offs accepted, and that, the report notes, is what competitors cannot easily copy. Data "becomes an advantage only when" the firm captures it at the point of creation, structures it for reuse and retains "the rights to learn from it over time". The learning loop it describes, estimates connected to outcomes, schedules to real progress, design choices to constructability, adds up to what it calls "a self-improving system".
On the invoice, the report is blunt about the arithmetic: AI will cut labour hours in an industry that mostly charges for them, and firms without new commercial models may give their productivity gains away. Its playbook item reads like a deadline, "Update commercial models in parallel with AI deployment, not after", and the destination it describes is a better business than the one being left: moving "from selling capacity to selling excellence with fewer surprises, more reliable schedules, lower risk, and better project outcomes". A Swedish reader is better equipped here than the report assumes. The standard consulting contract already permits outcome terms; the commentary to ABK 09's fee chapter opens for incentive- and bonus-based fees tied to agreed goals, results or performance. Practice is another matter, the auditor of the largest Nordic consultancy described the majority of its projects as running on löpande räkning in its 2020 audit report, while the contract forms on the entreprenad side, fixed price and unit price under AB 04, price commitments rather than hours. The gap between what the contract allows and what the sector does is the report's point, in Swedish translation. And the buyer's version of the same advice is a test for the next vendor meeting: if the pitch is finished outcomes and the price is per seat, the pitch and the price disagree.
On where the tools come from, the report's history lesson is short: "Historically, the largest technology shifts in AEC, such as computer-aided design, BIM, cloud platforms, and project-management software, came from outside the industry." It expects the pattern to repeat, notes that AEC firms have historically struggled to build and scale software, and points to recent acquisitions of AI companies by established firms. The names are easy to supply. In November 2025, AECOM bought the Norwegian startup Consigli, whose product is described as an autonomous engineer, and made its founder head of AI engineering. In September 2025, Rejlers signed on as one of the first Swedish pilot customers of Endra, whose software generates installation design from an architect's model. One bought the queue, one bought a place in it; neither tried to build the engine at home. The report's sidebar has advice for both: the parent should behave "like a supportive customer and growth partner, not a bureaucratic owner."
The keep-list
Read as a buyer, the most useful pages are the warnings about what not to give away. The first is accountability. "While AI can make many tasks easier to perform, it does not reduce the need for accountability," the report says under a heading of its own, and its corollary is generational: much of what is being automated is what junior staff have used to build "experience and judgment". A Swedish survey gives that warning a time-stamp. In June, a month before the report, Innovationsföretagen found that 78 percent of member firms said AI had not taken over tasks previously done by junior engineers, and eight in ten were still hiring new graduates. Read alone, that is reassurance. Read next to the report, it is a clock: the near-term list is the junior layer, and the firms hiring into it this autumn are hiring people whose first tasks will change under them. A firm that automates the drafts owes its juniors a deliberate route to judgment, structured review, early responsibility, real decisions. Done deliberately, the trade points upward: when drafting no longer fills the first years of a career, the work the title promised can arrive earlier in it.
The second is the decisions. The middle item of McKinsey's three names where decisions get made rather than how documents get produced, and in this industry the difference is structural. The production method is common property: Eurocodes, AMA, the calculation methods, the same shelf at every firm in the market, which is exactly why a machine can be taught to run it. What separates firms is which alternatives they choose and what they are willing to sign. A firm can hand over the producing and keep the deciding; it cannot buy the deciding back once it has moved to a partner, a platform or a client. The report's word for that outcome is ceding.
The third is about data, and it lands on familiar ground. "Vendor data rights are another advantage that AEC leaders consistently underestimate," the report says: as capabilities improve, vendors are "pushing harder to access project data and the rights to learn from it and reuse it to create new products". A Swedish consultancy already owns the vocabulary for this. ABK 09's chapter on rights grants the client use of a delivered result "endast för det med uppdraget avsedda ändamålet", only for the assignment's intended purpose, and the industry's stated reason is the one McKinsey now hands the whole sector: consulting is built on carrying knowledge and experience from project to project. The report's tests, whether you can move your data, keep it separated, and leave without losing it, are the same logic applied to a new category of supplier, and worth running, though an exportable file is the floor rather than the whole answer; files travel more easily than the working method around them. The question underneath is the one the report keeps circling, and it belongs in the next contract round: who may learn from your projects, and on whose terms.
The window is concrete: the report's near term is the first eighteen months, inside a single budget cycle, and the autumn it opens onto is already scheduled. A national plan is moving into procurement. The sector's largest client is under government orders to report on its own productivity by February. Six consultancies in ten expect their order books to grow. That is the part of the report that most deserves a board's attention: it describes a sector whose demand is outrunning its capacity to deliver, and it says plainly that firms which move early "may be best positioned to rewire their businesses to unlock value from AI". Held firmly, the keep-list is what makes moving fast safe. When the eleven pages reach your board agenda, the summary will already be in the room; the position worth bringing is the buyer's: one deliverable to delegate whole this winter, a route to judgment for the juniors who used to draft it, and a vendor contract whose data terms you read before signing. The window is eighteen months. The eleven pages take an afternoon.
Get news and articles in your inbox.
McKinsey's near-term list is the work Yesper already delivers end-to-end: whole deliverables from the documents where a project's cost and quality are decided, with the decisions and the signature kept by your engineers, and your data yours. Book a demo.
Book demo
