Brian Potter - Future of Construction, Ugly Modernism, & Environmental Review

Dwarkesh Patel introduces Brian Potter as an engineer and the author of the Construction Physics blog, which examines how the construction industry works and why it has been slow to industrialize and innovate. His opening question concerns The Line project in Saudi Arabia. Potter first praises the willingness of Saudi Arabia and other Middle Eastern countries to pour huge sums into ambitious infrastructure, something he says is rare in the modern world. China does it in large amounts and a few minor places do too, but few countries attempt projects this massive. He then draws a sharp line between the ambition and the plan: "the actual project is clearly insane and makes no sense."

His objection is about layout. Cities grow in two dimensions, he says, and a one-dimensional city is the worst possible arrangement for transportation because it maximizes the distance between any two points. He sees no real benefit unless there were extremely fast point-to-point transport, perhaps a bullet-train setup. Even if one wanted an entirely enclosed structure, which he also doubts makes much sense, a cube would save a great deal of material and effort, and he jokes that he would be more interested in the cube than the line. He says he would be surprised if The Line is ever built.

Asked why a cube matters, Potter explains that many economies of scale are geometric: as something grows, its volume increases much faster than its surface area. For enclosed things such as tanks or pipes, cost per unit transported falls with size. The same logic applies to buildings, because the exterior wall assembly is burdensome, complicated and expensive, so a building with a large floor plate gets more area per amount of exterior wall. Climate control is also a function of exterior surface. He is unsure, however, that this carries over to a single giant enclosed city, since such a city generates so much waste heat that it may be harder to pump out, as in the urban heat island effect. He says he is reaching back to college physics classes and is not sure of the actual mechanics.

Patel relays Scott Alexander's point that The Line is supposedly designed to save space and fuel, even though space and fuel are the one thing Saudi Arabia has plenty of, and asks what the thinking was. Potter's sense is that some committees have had success with large, impressive construction projects that attract attention by virtue of size, and he cites a great deal of what has been built in Dubai and a giant clock tower in Jeddah. They may expect a return simply from building the biggest thing on some axis, so he thinks they are optimizing for big and impressive without digging further. He adds a framework he thinks about often, the garbage can theory of organizational decision-making. In that view, organizations do not reach decisions through any particular rational process. When a problem arises, they reach into a garbage can of potential solutions, and whatever comes out becomes the decision, regardless of how much sense it makes. Academics invented the theory to describe decision-making in academia, but Potter applies it to big bureaucracies and governments, especially where decision-makers are disconnected from technical knowledge.

Patel turns to Potter's earlier eBay arbitrage business in designer clothes. Potter says he ran it seven or eight years ago, growing out of a hobby in high-end men's fashion that he admits is a strange one for an engineer. The hobby centers on finding cheap designer items, since buying new is overwhelmingly expensive, and he says nice clothes can be had cheaply on eBay or in thrift stores if you know what to look for. Well-made clothes last a very long time, so something worn for ten years could still be fine.

The tedium of eBay is what led to the project. Finding good items meant manually searching many brands, filtering out the obviously bad ones, and searching for the typos that sellers reliably put in titles. Potter built a simple web app that searched the nice brands and their misspellings, filtered out junk, and let users browse what was left. He added an affiliate system so he received a share of sales made through it. The site ran for a few years and was reasonably successful, and it was one of the first things he did that got real traction online. It never paid in proportion to the effort of maintaining it, so he stopped updating it as he moved away from the hobby.

Patel says he found it through a friend, Connor Tabarrok, who was with Potter at the Oxford Refugees Conference, and notes that finding other information about Potter online was hard because he had kept his anonymity. On the finances, Potter says the site made maybe a few hundred dollars a month for a few years. It was only ever a side hobby business, so per hour of effort the return was very low: "Pennies to an hour or something like that."

Patel summarizes a theme of Potter's post: the construction industry is plagued by lossy feedback loops, weak economies of scale, regulation and costly mistakes. He asks whether this is typical of many industries or unique to construction. Potter answers that many of the individual factors are not unique. Construction is highly regulated, but not necessarily more than medical devices or jet travel, or probably even cars, which have a whole vat of performance criteria to meet. Land use is another factor people raise, but many things in this category do not share the structure of the construction industry. In his view the explanation is the interaction of all these effects.

The factor he thinks is underappreciated is that a building's systems are highly coupled. In a computer, the hard drive is somewhat independent of the display and the power supply; the parts are coupled, but independent people who may never talk to each other can build them and then assemble them into one thing. A building is not like that: "Every single part affects every single other part." He compares it in some ways to biology, so it is very hard to change something without disrupting something else. The reason is that a building's job is to create a controlled interior environment, so every system must run through and around the surfaces that create that environment. He concedes this is not unique, since a plane or an iPhone shares it to some extent, but he says the amount paid per unit of size or mass is quite low for buildings.

Patel asks whether transportation cost is the fundamental reason a building cannot be as specialized and modular. Potter says it is really about the way a building is. The electrical system cannot be a separate box that you pull out and replace with a new one, because it runs through the entire house. The same is true of plumbing, insulation and interior finishes, so there is little modularity in a physical sense.

Patel then relays a comment from Ben Kuhn (given later in the transcript as "Ben Koon", as transcribed), who noted that sequential dependencies and highly variable delivery timelines, two reasons given for slow innovation in construction, are also common in software. If so, why has software not stagnated in the same way? Potter says a similar structure shows up in anything that is project-based, or where you figure out what you are doing while you do it. He contrasts that with large-scale manufacturing, where a lot of time goes into working out exactly what is being built, designing it to be built, and doing the first runs, after which the process is tweaked to be more efficient. Tweaking continues, but the figuring-out is largely separate from the doing.

Potter continues the comparison with software by saying a project-based industry cannot build its process in advance: "You have to build your process on the fly." Best practices shape the work, and someone writing software or making a movie has a rough idea of how it will come together, but many unforeseen things come up. The key difference is scale. A finished software project can be deployed to 1,000, 100,000 or a million people, and a finished movie can be watched by 100 million. A building has no such multiplier, so there is no way to spend a lot of upfront time working out how it should go that would be justified by its size. Potter says he dug up references showing that software projects have long tails, sometimes going massively over budget and often never being completed, which he finds shocking. The economics differ slightly because of how many people can use the product afterward.

Patel adds the famous software law that a project takes longer than expected even after accounting for that fact, and Potter identifies it, with some uncertainty, as Hofstadter's law.

Patel asks whether the lack of zero-marginal-cost scaling limits the size and number of construction startups. Potter says a construction software company can scale like any other software business, but physical products are much harder, and the absence of zero marginal cost has hurt many startups, not only construction ones. Construction is particularly brutal because margins are so low, and the empirical fact is that trying a more efficient way of building does not necessarily let you build cheaper and earn better margins. He draws on the startup he used to work at, Katerra, whose business model assumed that building in big factories would bring huge economies of scale and lower costs, allowing it to recoup the billions being invested. His verdict: "The math just does not work out." You generally cannot build cheaply enough to recoup such large upfront costs, and many businesses have been burned this way.

The most success in prefabrication, he says, is at the higher end of the market, where margins are higher. Higher quality is also easier to achieve in a factory than in the field. Houses built to a very high level of energy performance, for example, need very good air sealing, and these tend to be built with prefab and other factory methods because that is physically harder to achieve on site.