"AI and robots will automate construction, so buildings get cheap" is the shakiest leg of the [[Should you diversify into real estate in the age of AGI?|real-estate-in-AGI]] thesis. It bets against the single best-documented anomaly in the productivity literature, chronicled by Brian Potter ([Construction Physics](https://www.construction-physics.com/)): **construction costs rarely fall, and construction productivity has been flat-to-declining for ~60 years** — straight through CAD, power tools, nail guns, and cheap global materials. ## The base rate While the overall economy roughly *tripled* output per worker from 1970–2024, construction labor productivity is virtually flat; on some measures it has *declined ~50% since the mid-1960s* (Goolsbee & Syverson 2025, via Potter's [Trends in US Construction Productivity](https://www.construction-physics.com/p/trends-in-us-construction-productivity)). [McKinsey](https://www.mckinsey.com/capabilities/operations/our-insights/reinventing-construction-through-a-productivity-revolution) puts global construction productivity growth at ~1%/yr vs. 3.6% for manufacturing. This is a 60-year record of *not* industrializing — see also [[Construction costs have risen 15 percent since the pandemic]]. Real costs went *up*, not down. ## Why it resists automation - **"Bubba and his truck."** From Potter's [Katerra](https://www.construction-physics.com/p/another-day-in-katerradise) years: capital-intensive automated production kept losing to low-overhead contractors using manual labor and little else. Automating construction means deploying expensive, inflexible equipment to duplicate what cheap, adaptable labor does on a variable site. - **Low "idiot index."** Homebuilding's ratio of output cost to raw-material cost is only ~2 ([Economies of Scale](https://www.construction-physics.com/p/where-are-the-economies-of-scale)). Automation and scale make money by compressing the gap between input and output cost; in construction that gap is already thin, so there's little fat to render. - **Bespoke, tightly-coupled integration.** Every part affects every other part on a site-specific build, defeating the repetition automation feeds on. - **Fragmentation.** 65,000+ homebuilding firms; the top 100 control <50% of the market. The "big" builders (Lennar, D.R. Horton) subcontract ~all actual construction and act as financiers — nobody has the scale or R&D budget to industrialize. ## Prefab and robots keep failing — for structural reasons - **Katerra burned ~$2B** and died of "a lack of product-market fit — it scaled massively before having a product people wanted." Even **manufactured housing** — single national HUD code, factory repetition — shows *minimal* economies of scale: small plants match or beat the giants, because transport is cheap-per-mile relative to bulky volume (~350-mile radius) and factory fixed costs become a liability the moment demand wobbles ([The Prefab Pivot](https://www.construction-physics.com/p/the-prefab-pivot)). Potter has publicly recanted his own earlier "factories will be cheaper" belief. - **Robotic bricklayers** (SAM, Hadrian X) have been in development since 2006/2015 and still do a trickle of projects; SAM can't turn corners or finish joints, so humans follow behind ([Where Are The Robotic Bricklayers?](https://www.construction-physics.com/p/where-are-the-robotic-bricklayers)). - **3D printing** only addresses the shell. [Icon's printed walls](https://www.construction-physics.com/p/what-progress-has-icon-made-on-3d) are still 2–4× a conventionally framed wall, and windows, MEP, finishes, and waterproofing all remain. ## The composition trap Automation, prefab, and 3D printing all attack the **structural shell** — but that's not where the cost-and-labor mass sits. Within a home's construction cost ([NAHB 2024](https://eyeonhousing.org/2025/01/cost-of-constructing-a-home-in-2024/)), the structural framing is ~17%, while **interior finishes (24%) and mechanical/electrical/plumbing rough-ins (19%) dominate** — exactly the fiddly, trade-specific, hard-to-robotize work. Potter's deepest point: the real bottleneck is the *operating system* around the build — permitting, fragmentation, site-specific design, liability — not the hammer. So the honest "this time is different" bet isn't "robots swing hammers"; it's "AGI dissolves the regulatory/design/coordination layer." That's a far higher bar. ## What would change my mind One channel bypasses all of the above: **materials are ~half of construction cost, and cement and steel are energy-intensive** — so radically cheap energy would lower material costs even if labor never automates. That's the version of "AI makes building cheaper" with real legs. But it requires energy to get *cheaper*, and right now AI data-center demand is pushing it the other way ([[Energy costs]]). If delivered industrial energy starts falling hard, revisit this note. Source: Brian Potter, [Construction Physics](https://www.construction-physics.com/). #Question: does the energy-→-materials channel ever overwhelm the 60-year stickiness base rate? Tracked in [[Energy costs]].