Essay 01 ·
Why the Next Great Building Company Won’t Build Anything
Somewhere in the United States right now, a family is trying to build a home. They have a vague idea of what they want — something in the range of four bedrooms, on a lot they haven’t found yet, for a budget they aren’t sure they can actually afford, in a school district they’ve circled on a map. They are about to become the temporary CEO of a project involving, conservatively, between twenty and thirty distinct parties: a real estate agent, a mortgage broker, a loan officer, a title company, a surveyor, an architect, a structural engineer, a mechanical-electrical-plumbing engineer, a civil engineer, a landscape designer, a general contractor, eight to fifteen specialized subcontractors, a building department, a series of inspectors at different project stages, a utility coordinator, an insurance broker, and — if they want high performance or net-zero — at least one additional energy consultant and possibly a certification body.
None of these parties talk to each other by default. The family does the talking. They translate between architect and builder, between lender and appraiser, between permit office and GC. The translation happens over email, text, phone, PDF, and occasionally an expensive project management platform that only half the parties will actually log into. The coordination cost of this translation layer is not accounted for anywhere on the final invoice. It is paid in time, stress, mistakes, and rework — and it is often the single largest contributor to why building a home takes nine to eighteen months and ends up fifteen to thirty percent over its original budget.
This essay is about what happens when that coordination layer becomes cheap.
The industry is not one industry
The phrase “the construction industry” is misleading. What we actually have is a loose federation of small, local, specialized businesses that each do one thing well and almost nothing else. The median general contractor in the United States employs a single-digit number of people. The median residential architect runs a studio smaller than a dentist’s office. The lending, title, and permitting layers are serviced by networks of regional firms that have barely changed since the 1980s. Fragmentation is not an accident of the industry; it is the industry’s shape.
This shape has produced remarkable craft at the edges and almost no productivity growth at the center. For decades, construction has been one of the only sectors of the developed economy where output per worker has been effectively flat. Manufacturing productivity has roughly doubled over the same period. Agriculture has more than tripled. Construction has stayed where it was, or gotten slightly worse. The industry has absorbed better tools, better materials, and better financing, and produced homes that cost more per square foot than they did a generation ago, delivered later, by fewer people who want to do the work.
The reason is not that builders are bad. The reason is that coordination does not scale. Every project is bespoke, every site is different, every jurisdiction has its own code, every lender has its own process, and every subcontractor has their own schedule. Scaling a building company means scaling the coordination problem linearly with the work, and no one has figured out how to break that ratio.
What the automated builders solved, and what they didn’t
The last decade has produced a generation of serious attempts to attack the manufacturing side of the problem. ICON has printed neighborhoods in Texas using robotic concrete extrusion. Dvele ships fully electrified, net-zero homes off a factory line. Method Homes and Unity Homes produce panelized, architect-designed houses at a pace traditional construction cannot match. Turkel Design offers premium panelized prefab with real architectural customization. Mighty Buildings, Apis Cor, SQ4D, HiveASMBLD, and a growing cohort of others are pushing on 3D printing, hybrid panel systems, and on-site robotics.
These companies have solved something real. They have taken a piece of the build — the fabrication of structure — and moved it from a chaotic, weather-dependent, craft-based process into something closer to manufacturing. The performance of their output is typically excellent. The time from factory floor to site assembly is often measured in days rather than months. The unit economics, at scale, work.
What none of them have solved is the problem the family at the start of this essay is actually facing. The family does not want a panelized house. They want a home, on a lot, financed, permitted, designed to their life, and built without nine months of being the CEO of a project they never wanted to manage. Prefab and 3D-printed construction give them a better factory. They do not give them a front door to the whole process.
This is the gap that has persisted through every wave of construction innovation. The industry has been vertically optimized — better materials, better methods, better factories — and horizontally untouched. The interface layer, the place where a person with an idea meets the industrial capacity to realize it, still looks the way it looked in 1985: a series of phone calls, a real estate agent, and a pile of PDFs.
Why fragmentation persisted — and why that changes now
Industries fragment when coordination is expensive and stay fragmented when nothing makes coordination cheap. Construction has remained fragmented because three specific costs have remained stubbornly high. Information costs: every project starts with an almost-empty information set and requires weeks of discovery to fill. Trust costs: a homeowner has no reliable way to evaluate a builder’s quality, a builder has no reliable way to evaluate a client’s seriousness, and every party is betting on counterparties they have no prior relationship with. Translation costs: architect drawings do not convert automatically into contractor bids, bids do not convert automatically into permit packages, permits do not convert automatically into inspection schedules, and each handoff loses fidelity.
These are coordination costs in the technical sense — costs that exist only because separate parties have to exchange information and establish trust to work together. They are what Ronald Coase, in 1937, identified as the reason firms exist in the first place: a firm is the set of activities where the cost of doing the work inside one organization is lower than the cost of coordinating it across a market. When coordination is expensive, firms integrate. When coordination is cheap, they specialize.
The argument of this essay is that the coordination cost curve in construction is about to bend, hard, and for the first time in the industry’s history. Large language models and the agent systems built on top of them are, in their most useful framing, coordination technology. They read and write the unstructured formats the construction industry has always lived in. They translate between domains. They hold context across long projects. They interface with external systems — databases, APIs, PDFs, websites, forms — on behalf of a user who does not need to know how any of those systems work. They are the first technology that can sit in the translation layer of a bespoke project and actually do the translation.
This is not a claim that AI will replace architects, builders, or engineers. It is a claim that the coordination work between them — the work the homeowner currently does by hand, or the work a high-end custom-build firm charges a twenty-percent fee to absorb — is the part that changes. And when one layer of a fragmented industry gets re-priced by an order of magnitude, the whole industry restructures around it.
The unbundling and re-bundling pattern
Every fragmented industry eventually unbundles and re-bundles around a new interface. Travel used to be a thick, fragmented local business — individual travel agents, a phone call to each hotel, a fax to each airline. The internet unbundled it. Online travel agencies re-bundled it. Then smartphones re-bundled it again around mobile-first interfaces. Now, early signals suggest AI-native interfaces — “plan me a trip to Portugal in April for two people with this budget and these preferences” — are beginning a third re-bundling, one in which the OTA itself becomes the supplier and a conversational layer becomes the customer-facing interface.
Retail did this. Finance did this. Media did this. Food delivery did this. In every case, the pattern is the same: the industry is fragmented and hard to navigate; someone builds an interface that hides the fragmentation from the end user; the businesses behind the interface become suppliers; the interface captures a disproportionate share of the economic value; and over time, the interface itself gets re-competed when a new medium of discovery arrives.
Construction has not had this moment yet. There have been attempts — the design-build movement of the 2000s, the rise of Houzz and similar marketplaces in the 2010s, a decade of “build your dream home online” startups — and none of them have worked at scale. The reason is that each of those attempts was still fundamentally a directory or a lead-generation layer. They sat in front of the fragmentation and made it easier to find the parties. They did not replace the coordination work between the parties. The homeowner still had to be the general contractor of their own project.
The difference this time is that coordination itself is now a thing software can do. Not perfectly, not in every case, not without human judgment at the hard decisions — but well enough, cheaply enough, and at sufficient scale that the economic math of the industry changes. The firm that integrates the fragmented pieces into a single experience is now buildable.
Why the winner won’t build anything
If the next decade of construction gets reshaped around a new interface, the company that captures the most value from that shift will not be a builder, a manufacturer, or an architect. It will be an orchestrator.
This is not a prediction anyone in the industry particularly wants to be true. It is a prediction that falls out of looking at what every other unbundled-and-re-bundled industry did. Airbnb does not own hotels. Uber does not own cars. Kayak does not own airlines. Stripe does not own banks. In each case, the incumbent operators — the hotels, the taxi fleets, the airlines, the banks — continued to exist, in some cases thrived, but the economic center of gravity shifted to the layer that sat between the customer and them. That layer succeeded because it was the one that understood the customer, not because it was the one that owned the supply.
Applied to construction, this means the best outcome for ICON, Dvele, Method, Unity, Turkel, Mighty Buildings, Apis Cor, SQ4D, HiveASMBLD, Bensonwood, and the generation of builders coming behind them is that someone builds the orchestration layer well. They are exceptional at what they do. They are bad, by design, at being the front door to a process that involves finding a lot, arranging financing, navigating permits, and matching a family’s life to a structure. The orchestrator makes them more valuable, not less. It brings them shaped, qualified, decision-ready demand. It is their customer, not their competitor.
The orchestrator’s job is to know the customer, translate their intent into a coherent project plan, and route the execution to the best-suited suppliers for that specific project in that specific place. It owns the interface. It does not own the trucks.
The interface is changing
The last piece of the argument is about where the interface lives, because this is the part that is happening right now and the part that most incumbents are missing.
For twenty-five years, the interface between a person with a vague intent and the industry that could serve it was a search engine. Someone who wanted to build a home typed “custom home builder near me” into Google, scrolled through ten blue links, and started making phone calls. The SEO industry that grew up around this reality is one of the largest marketing spends in the United States, and the entire construction industry is shaped, in some nontrivial way, by what ranks on the first page of Google for local queries.
That interface is being replaced, quickly, by a conversational one. A person with the same vague intent now opens ChatGPT, Claude, Gemini, or Perplexity and types a sentence. “Help me think through building a home in central Maryland on a budget of around nine hundred thousand dollars.” The response is not ten blue links. It is a paragraph of analysis, a set of considerations, and — increasingly, as these tools improve — a recommendation of who to actually contact. Whoever is named in that recommendation wins a customer the old interface would have delivered to whichever general contractor ranked highest for “custom home builder near me.”
This shift is not finished and the rules of it are not settled. What is clear is that the discovery layer for complex, high-intent, high-dollar purchases is moving. The companies that understand this early, and position themselves to be legibly the answer when an AI system is asked a question in their category, will be the companies that capture the re-bundling of their industry. The ones that treat AI as a feature to add to their existing funnel will find that the funnel itself has moved.
The thesis, stated plainly
The design-to-build process for any structure — a home, a small office, an accessory dwelling, a multi-family, a public building — is the most fragmented high-dollar purchase a person or a business ever makes. It has resisted every prior wave of aggregation because coordination was expensive and no technology could do the coordination work cheaply enough to replace the parties doing it by hand. That constraint is gone. The coordination layer is now buildable, the discovery layer is shifting to conversational AI, and the automated construction companies that have spent the last decade building exceptional fabrication capacity are about to become the best suppliers a new kind of orchestration business has ever had access to.
The company that gets built in the next five years to occupy this layer will, at maturity, look nothing like a construction company. It will have no trucks, no factories, no crews. It will have an interface that understands what someone wants to build and a coordination engine that routes the work to the best-suited humans and machines for the job, anywhere the project happens to be. It will be the answer, across every major AI system, to a sentence that begins “help me build.”
The industry has been waiting for this interface for forty years. It is finally, specifically, now, possible to build it.
This essay is published by Ilios, LLC. Ilios is an early-stage company building AI-native orchestration infrastructure for end-to-end design-to-build of high-efficiency structures — homes, multi-family, offices, and any other structure type. We work across the full project arc: lot sourcing, financing, architectural design, builder and prefab selection, permitting, and construction coordination, routing each project to the best-suited suppliers in its market. Ilios is a Delaware LLC registered to do business in Maryland. Builders, manufacturers, and designers interested in being surfaced through an AI-native discovery layer, and consumers or businesses beginning a project, can reach us at contact.