top of page

Can Algorithms Design Buildings?

After decades of unsuccessful attempts to generate building layouts automatically, a spate of companies has suddenly proven it possible.

Humans have been trying to harness the power of computers to automatically generate building designs for decades. Like turning lead into gold, it seemed like a foolhardy endeavor that consumed many hopefuls. But after years of tepid results, a number of companies are finally cracking the alchemy of algorithmic space planning.

On a lush street lined with trees, bike lanes, and modest Dutch townhouses in the city of Alkmaar, 20 miles north of Amsterdam, a vacant, overgrown site is about to be turned into a housing development. The mastermind behind one of the proposed development schemes is The Living, a New York–based research group founded by David Benjamin and acquired by Autodesk in 2014. The prospective project developer, the Van Wijnen Groep, had seen how The Living employed generative algorithms to create Autodesk’s MaRS Office, in Toronto, and believed a similar process could generate housing development plans.

Working with the parameters of the Alkmaar site, The Living developed an algorithm that could—with the supervision and guidance of a designer—lay out, evaluate, and refine buildings. Meanwhile, the Van Wijnen Groep’s own designers created competing plans using their conventional processes and rules of thumb. In comparing the results, The Living’s project lead Lorenzo Villaggi says, the manually created designs “took an approach that was known to work based on past experience, but the algorithmic solutions moved beyond what you would typically think of.”

Because the software could rapidly churn through essentially countless variations, it allowed the designers to explore the less obvious solutions that better balanced the developer’s need for profit against the community’s need for light, thermal comfort, energy conservation, and amenities. As Benjamin tells me, “Some people seem to have this assumption that sustainability costs more, but in this case we can do both.”

A few months ago, a real estate agency began selling speculative houses on the Alkmaar site. When shown the plans, Benjamin and Villaggi weren’t able to immediately confirm whether the residential units were a product of their algorithms. After some investigation, it turns out that while the Van Wijnen Groep has been applying The Living’s methods to other projects that would eventually be built, another developer had won the rights to build on the Alkmaar site.

Still, the uncertainty about the origin of the designs is telling. Historically, computational designers have flaunted their digital underpinnings with superfluous curves and other flourishes, but the schemes produced by The Living’s algorithms are largely indistinguishable from that of a human designer. Looking at one of their schemes, you can’t see the algorithm; that is, you can’t tell that every angle, measurement, and configuration was tested thousands of times until everything was just so. Passing this type of Turing test feels like a new step forward for computational design.

The Living is but one of several firms that have recently and successfully employed algorithms to lay out buildings. Other examples include Higharc

, which is creating software to automatically lay out residential designs; Spacemaker, Archistar, and TestFit, which are all doing the same for commercial real estate; and WeWork, which is developing tools to automate office layouts (disclosure: I managed this initiative at WeWork).

TestFit CEO and co-founder Clifton Harness says that its software has been used to design 10,000 residential apartment buildings in the past two years. He’s quick to add that most of these projects aren’t built, but rather courtesy of efforts by developers—TestFit’s primary users—who are exploring what is viable on their properties with the platform before engaging an architect.

Harness himself spent the first two years of his career laying out residential buildings for a developer: two per week, week after week. After producing close to 200 designs manually, he approached his fellow University of Texas alum Ryan Griege to see if they could devise a way to automate his job. Working nights and weekends, the pair set about developing a suite of layout algorithms. In 2016, they quit their jobs and started building a business around the technology. Griege is now the chief technology officer of TestFit.

While tech-savvy architects have long employed computers to help design discrete, one-off building components, such as the details of a curved façade or the size of truss members, creating an entire building algorithmically remained elusive.

The early work on algorithmic layouts is perhaps best documented in Nigel Cross’s 1977 book The Automated Architect (Pion, 1977), which walks through the process of computationally generating floor plans. Cross started with the belief “that computers might produce designs that are somehow better than designs produced by humans.” Now in his late 70s, Cross tells me via email that by the time he came to write the book, he “didn’t think that automating architecture was a good idea.” He spent the rest of his career studying and celebrating human designers. Looking back, he writes, “For me, the value of asking the question, ‘Can a machine design?’ is that it begs the corollary question, ‘How do people design?’ ”

A lot of this early work failed to take off, mainly because most architects at the time didn’t employ computers, and the few who did were struggling even to produce drawings. These early setbacks didn’t dissuade generations of starry-eyed newcomers—including the former dean of MIT’s School of Architecture and Planning William Mitchell and the Google-funded startup Flux—from trying their hand. But after decades of effort, none of this work was widely used.

So what changed? Why have many companies succeeded in algorithmically generating floor plans in the last few years?

Anthony Hauck has led a number of efforts in this area, first at Autodesk where he originated Project Fractal and more recently as president of Hypar, where he and Dynamo Studio originator Ian Keough have been working to create a universal computing environment for the AEC industry. Hauck tells me that “technological capabilities are great,” but the real change has been “a cultural shift brought about by actual successes in the field.”

Both David Benjamin and Clifton Harness give a similar explanation, attributing their progress and—dare I say—success to a steady cultural change rather than a major technological breakthrough. Benjamin says that when he started this work a decade ago, computational design was still exotic and people took a lot of time to warm up to it. Fast-forward to today and everyone is “familiar with using computational models for geometry.”

At the same time, the economy is buzzing and unemployment among architects is low. Money is more readily available for risky projects and an incentive exists to reduce labor costs through automation. Adding to the pressure, architects and developers are observing the impact of automation on other sectors, from accounting to medicine, and wondering what is in store for the architecture industry.

One possible outcome is that, yes, these tools replace architects by doing aspects of their jobs faster and more efficiently. Even in this extreme scenario, the threat is fairly minimal since architects only spend a small part of their day on layouts, with this task often falling on interns and recent grads. Furthermore, most of these algorithms require a driver behind the wheel, so to speak, and technologically it seems unlikely companies will be able to remove the steering wheel anytime soon.

The much more likely scenario is that, rather than replacing architects, these tools make architectural expertise more widely available. As Benjamin points out, a large number of buildings are already designed without architects—such as the rows of tract housing in the suburbs—and these automation tools may lower the cost of design to a point where “architecture firms could chip away at these markets.”

Hauck agrees. “There is an enormous amount of latent demand,” he says. “If this expertise was as widely available as WebMD, Rocket Lawyer, and TurboTax, then maybe we’d get a better built environment.”

The Alkmaar project is a case in point. Whereas one might expect the application of The Living’s sophisticated algorithms on landmark projects by name-brand architects, the firm's computational efforts are deployed at an everyday housing development on an overgrown site in the outskirts of Amsterdam.


4 views0 comments


bottom of page