The strongest productivity claim for AI in infrastructure design is not about speed. It is about how many alternatives an engineer can afford to evaluate before the design hardens. Today that number is one, sometimes two. Toyota showed what happens when it rises.
The claim
You have heard the pitch: AI makes engineering ten times faster. We do not make that claim. It rarely survives contact with a real project, and it aims at the wrong prize anyway. The claim that does survive scrutiny is smaller, more precise, and worth more: one more design iteration per project, at the same fee.
An iteration here means an evaluated alternative. A second bridge alignment carried through quantities and cost, not just sketched. A foundation concept actually checked against the ground investigation. A system choice with numbers behind it instead of a hunch. The promise is not that projects get cheaper because engineers move faster. It is that a fixed fee buys more evaluation: more alternatives examined before one of them becomes the design.
The evidence
The best evidence that iterations pay comes from an industry that measured it. In the early 1990s, researchers at the University of Michigan studied why Toyota developed better cars faster than its Western competitors. The answer surprised them. Toyota did not pick a concept early and refine it, the way its rivals did. It kept whole sets of alternatives alive in parallel, tested them against each other, and committed late. The researchers published the finding in Sloan Management Review under a title that carries the entire argument:
The Second Toyota Paradox: How Delaying Decisions Can Make Better Cars Faster.
The paradox is that exploring more alternatives, for longer, ought to be slower and more expensive. It was neither. By Ward's estimate, set-based development made Toyota's engineering roughly four times as productive as its competitors', and the cars came out better. The alternatives were not waste. They were how Toyota found out which decision was right while it was still cheap to be wrong.
The constraint
Infrastructure design runs on the opposite logic, and not because engineers prefer it. A typical design commission affords one iteration, sometimes two, because each one costs weeks. Behind every alternative sits a chain of documents: quantities to recalculate, technical descriptions to update, requirements to trace through once more, estimates to rebuild. All of it produced and checked by hand, inside a fixed fee. So the second alternative dies in a meeting. It survives as a sentence in the study ("a western alignment was considered but not pursued") instead of as a costed option.
Now change one number. When the document chain around a design decision runs at machine speed, an iteration costs an afternoon instead of three weeks. Nothing else about the engineering changes: the same engineer makes the same judgements and signs the result. But the alternative gets evaluated instead of argued about. At that price, running three alternative studies is no longer a luxury a project manager has to defend. It is the obvious way to work.
The leverage
Where the extra iteration lands matters as much as having it. In 2004, Patrick MacLeamy, then chief executive of the architecture firm HOK, drew a curve that has followed the industry ever since: the ability to influence cost is greatest at the start of design, while the cost of making changes grows exponentially as the project advances. A revision at concept stage costs a meeting. The same revision during detailed design costs a redesign. During construction, it costs a claim.
The MacLeamy curve is usually read as an argument for shifting effort earlier. It is equally an argument about iteration: cheap, early alternatives are the mechanism by which expensive, late changes are avoided. A design that has survived three evaluated options in the planning stage carries fewer buried surprises into detailed design and construction. One more iteration where changes cost one, so there are fewer changes where they cost a hundred.
The payoff
There is a second payoff, one level above the single project. Infrastructure portfolios suffer from a statistical winner's curse: when projects are chosen on uncertain early estimates, the chosen ones are systematically the underestimated ones. We wrote about that mechanism in where overruns are born. More iterations attack it at the root. An estimate that rests on three evaluated alternatives is less uncertain than one resting on a guess, and less uncertainty means less systematic underestimation among the projects that win funding.
Note what the claim is not. It is not that design gets faster and therefore cheaper; speed alone buys nothing if it is spent producing the same single option sooner. The claim is more evaluation per krona: the same fee, the same engineers, the same responsibility, and more alternatives genuinely examined before commitment. Toyota called it a paradox that delaying decisions made better cars faster. It stops being a paradox once you can afford the alternatives. If a study on your desk has an option that was set aside for lack of time, that is the iteration this argument is about. Run it.
Sources
Yesper is the AI civil engineer for construction and infrastructure. AFRY, COWI, NRC Group and other Nordic firms use it to halve the time on a study, rerun calculations in minutes, and catch errors that would otherwise slip through. Get in touch if you'd like to see what it can do for you.
Book demo
