Your Contractor Opened the Wall and Found Galvanized Pipe, Knob-and-Tube Wiring, and a Beam That Wasn't on Any Drawing.

Forty-five thousand dollars. Monica and David Chen had three bids, a signed contract, and a start date for their kitchen renovation at that price. On day two, their contractor opened the wall behind the sink and called with the voice people use when the number is about to change: galvanized supply lines running to every fixture, knob-and-tube wiring looped through the ceiling joists, and a header spanning the old doorway that was undersized by two inches and not attached to the king studs on either side. None of it appeared on any drawing because no drawing of this 1948 bungalow existed, at least not one that reflected what seventy-eight years of previous owners had actually done to it.

Final cost: $68,000.

Nobody in the industry finds this surprising. According to data compiled from contractor surveys and claims records by DoJo Business in 2026, hidden structural problems appear in 40 to 50 percent of renovation projects. Outdated electrical or plumbing surfaces in 35 to 45 percent. Change orders hit 70 to 80 percent of all projects. Standard advice is to budget a 10 to 20 percent contingency on any renovation, and for homes built before 1985, add another 10 percent specifically for code-compliance surprises. An entire industry has normalized the ambush.

The Invisible Building

Architects have a quiet, persistent problem that predates AI by centuries. We draw what we can see and measure, and we infer the rest. When I walk through a 1920s Craftsman slated for a second-story addition, I can read the visible structure: the exposed joists in the basement, the roof pitch from outside, the electrical panel in the garage. But the walls are opaque. Between the plaster and the sheathing you might find balloon framing that runs unbroken from the sill plate to the attic, a fire hazard so severe that modern code forbids it, or sistered joists layered onto old ones in a pattern that makes no structural sense. I do not know until someone opens the wall, and by then the demolition contract is signed and the dumpster is in the driveway.

David Benjamin, Director of AEC Industry Futures at Autodesk, calls the aspiration "a kind of x-ray vision for buildings." His team, working with Arcadis, has built an AI system that combines laser scans, embedded sensors, old floor plans, and geospatial data to generate 3D models predicting hidden infrastructure inside existing structures. His key insight is counterintuitive: rather than demanding one comprehensive, pristine dataset, the system layers multiple imperfect ones. A floor plan from 1952 that shows the original plumbing layout. A laser scan that captures the current wall positions. A sensor reading that detects moisture behind the tile. Individually, each dataset has gaps, but combined through machine learning, they produce predictions that exceed what any single source could deliver alone.

"Every time AI makes a prediction, it has a sense of how certain or uncertain it is, but this information is rarely visible to the user." — David Benjamin, Autodesk

That uncertainty modeling matters enormously, because the tool does not claim to know what is inside a wall. It assigns probabilities: a 92 percent likelihood of cast-iron waste lines based on construction era, regional building practices, and sensor signatures, or a 67 percent chance that the original knob-and-tube was replaced based on the permit history and thermal scan. What an architect or contractor actually sees is a heat map of confidence, not a false promise of certainty, which means the pre-renovation assessment becomes a conversation about risk rather than a coin flip during demolition.

Castellana 66 and the Commercial Proof

One working proof of concept exists. Castellana 66 in Madrid, a 1990 office tower, underwent an energy retrofit guided by the Autodesk-Arcadis AI model. Arcadis Global Director of Architecture and Urbanism Mansoor Kazerouni described the core challenge: "Existing conditions aren't always clearly documented. You don't truly know what you'll encounter until work begins." Using the fused-data approach, the AI predicted hidden mechanical systems, insulation gaps, and structural constraints with enough accuracy to plan the retrofit before opening walls. Result: 55 percent reduction in annual energy use and 10,800 tons of prevented CO2 emissions.

Impressive, and also irrelevant to most homeowners.

Castellana 66 is a commercial building with institutional documentation, a facilities management history, and a client willing to pay for LiDAR scanning and sensor deployment. A residential kitchen remodel operates in a different universe. What passes for homeowner "documentation" is a disclosure statement from the previous sale, maybe a permit pulled in 1987 for a bathroom addition, and a vague memory from the neighbor who watched the roof get replaced sometime during the Clinton administration. Scanning technology that works on a Madrid office tower costs $2,000 to $5,000 for a residential deployment. On a $45,000 kitchen remodel, that scanning cost alone represents 4 to 11 percent of the budget, which is roughly equal to the contingency fund that the scanning is supposed to protect.

The Math That Should Change Somebody's Mind

Set the scanning cost aside for a moment and consider the downstream numbers.

If hidden conditions appear in 40 to 50 percent of renovation projects, and the average surprise costs $8,000 to $15,000 when it involves electrical or plumbing, then the expected cost of ignorance on any given renovation is roughly $3,200 to $7,500 (midpoint probability times midpoint cost). A pre-renovation AI scan that caught even half of those conditions would save $1,600 to $3,750 per project, which is within striking distance of the scanning cost itself. Not absurd as a break-even proposition. Merely premature.

Scanning costs need to drop by about 60 percent before the economics work for a typical residential job, or the accuracy needs to climb high enough that contractors can skip exploratory demolition entirely, converting a $2,000 scan into $5,000 of avoided rework and schedule delay. Neither threshold has been met, but both are plausible within five years, given that LiDAR sensor costs have fallen 90 percent since 2016 and residential-grade 3D scanning devices like the Matterport Pro3 already capture spatial data at $500 per session.

What You Can Do Right Now

Right now, the Autodesk-Arcadis tool is experimental. You cannot buy it or hire a firm that deploys it on houses. So what is a homeowner renovating a 1948 bungalow supposed to do with this information?

First, know the age of your home's systems. NFPA estimates that knob-and-tube wiring remains in roughly 30 percent of homes built before 1950, and HUD puts the number of U.S. homes containing lead-based paint at 37 million, which covers everything built before 1978. Galvanized steel supply lines, common in homes built between 1930 and 1960, have a functional life of 40 to 70 years. If your home was built in 1950 and nobody has repiped it, the plumbing is 76 years old. Nobody should be surprised when it fails during renovation.

Second, hire a pre-renovation inspection, and not a standard home inspection, which evaluates the house for sale. Pay for a scope-specific assessment where an inspector uses a thermal camera, a borescope, and (if budget allows) a moisture meter to probe the specific walls, floors, and ceilings your contractor plans to open. Cost runs $300 to $800, depending on scope, and it will not find everything, but it will find enough to prevent the worst surprises.

Third, demand itemized contingency in your contract rather than a lump 10 percent line. Separate line items for electrical, plumbing, structural, and environmental contingencies, each with a trigger condition and a maximum draw. This forces your contractor to think through the specific risks of your specific house rather than applying a generic percentage.

Limitations of This Analysis

Autodesk-Arcadis research has been demonstrated on one commercial building. No peer-reviewed study has validated the approach on residential structures, where construction methods vary wildly by region, era, and the whims of individual builders. That 40 to 50 percent statistic for hidden structural problems comes from contractor surveys, not controlled studies, and suffers from selection bias: contractors remember the bad surprises. My break-even calculation uses midpoint estimates from cost ranges that span factors of three or more, and the assumption that AI scanning accuracy will improve on a residential scale is an extrapolation from commercial data, not a demonstrated residential result.

Here is the strongest case against this technology: residential renovation is too varied, too small-scale, and too poorly documented for the data-fusion approach to work. A 1948 bungalow in Phoenix, built with desert-adapted methods on a concrete slab, has almost nothing in common with a 1948 colonial in Boston sitting on a fieldstone foundation. Training data required to handle that diversity does not exist yet, and assembling it would require scanning thousands of homes during renovation, a chicken-and-egg problem that no market incentive currently solves.

Buildings account for 40 percent of global carbon emissions, according to Autodesk's own research, and renovation in the United States alone exceeds $500 billion annually, per the Joint Center for Housing Studies at Harvard. What separates what we know about existing buildings from what we need to know to renovate them efficiently is not a niche problem; it is one of the largest sources of waste in construction. AI to close that gap exists in prototype, and the economics do not yet support it at the kitchen-remodel scale. But the math is moving, and the walls, for the first time in the history of this industry, are becoming slightly less opaque.