The built world is the largest underdigitized market on earth. Construction and the broader AEC industry — architecture, engineering, and construction — move trillions of dollars a year, and almost none of it runs on software the people doing the work actually love. The tools are bolted on, not built in. The judgment that matters still lives in the heads of engineers, the files of permitting agencies, and the routines of contractors who have done this for thirty years.
For most of the last two decades, that was a problem nobody could solve cheaply. Capturing expert judgment in software meant hiring the experts, encoding their reasoning by hand, and watching that encoding rot the moment a regulation changed. The economics never worked outside the largest enterprise accounts — which is precisely why the incumbents who did break through built real businesses, and real moats, around a handful of workflows.
What has changed is the cost of judgment itself. The frontier AI models are now genuinely capable of reading a set of plans, interpreting a regulation, drafting a permit application, and explaining their reasoning the way a competent junior engineer would. We are not speculating. We use these models every day against real civil engineering and permitting work, and they are good — good enough that the bottleneck has moved.
The bottleneck is no longer the model. It is knowing which problems are worth solving and having the domain expertise to tell when the answer is right. That is where we see the opportunity.
We see it most clearly in the regulatory layer — permitting, compliance, and the slow machinery of approval that sits between a good project and a built one. Permits are the part of the built environment everyone hates and no one has fixed. The information is public, scattered, and inconsistent; the process is human, repetitive, and expensive. It is the perfect surface for AI wielded by people who understand it.
We see it in civil and geotechnical engineering, where enormous amounts of skilled time go into analysis and reporting that follows known methods and known standards — work that an AI-native tool, supervised by a licensed engineer, can compress without compromising rigor.
And we see it across the contractor economy, where the back office still runs on paper, spreadsheets, and memory, and where a small amount of well-designed software changes how a business competes.
None of this is a thesis about AI replacing engineers. It is a thesis about leverage. The people who win the next decade in the built world will not be the ones with the best models — everyone will have those. They will be the ones who understand the work deeply enough to point the models at the right problems and to stand behind the output. We are civil engineers, builders, and software people who have done all three. We intend to build the tools that define this layer, and we are building them now.