Question: How many news articles, think pieces, and scholarly papers about AI have you read in the past few months? For me, it’s too many to count. AI is nearly as ubiquitous and annoying as Donald Trump. And the claims for it, as well as the fears aroused by it, appear to inspire two types of hasty, hysterical overreactions: AI will either usher in a brave new world of efficiency, scientific breakthroughs, and medical miracles (Digital Utopia) or enslave us in a creepy, canned feudalism ruled by robots and the oligarchs who own them (Dystopian Nightmare). As for our little corner of the world—architecture—what possible role will human designers have when the cost of producing drawings drops to pennies? Here’s a sliver of truth: We’re at the start of this explosion of computing power and, as a result, are stuck in a speculative vortex of hype, hope, fear, conjecture, and bullshit. And yet AI itself is real, if for no other reason than a huge chunk of the world’s economy is currently heavily invested in its future, however scary or dubious.

After talking to a number of people about AI and the future of architecture, experts who’ve spent considerable time pondering the uncertainties, the famous quote from the great screenwriter William Goldman came to mind: “Nobody knows anything. … Every time out it’s a guess and, if you’re lucky, an educated one.” 

The same almost certainly applies to predictions about AI and the future of architecture. Educated guesses might be the best we can do for the moment, since whatever we can say about AI right now (It can’t draw a damn floorplan!) might be out of date in a month, a week, a handful of days. Architect Randy Deutsch, author of The Agentic Architect: AI and the Resurgence of Practice (Routledge, 2026), has long grappled with the challenges of charting the progress of new technologies. “I’ve written seven books,” he says. “All of them involved design technology, all of them referenced AI. In order to extend their shelf life, I don’t mention specific tools by name, because by the time the book comes out, the tools have been usurped by other tools, they become menu items, they merge with other products, other tools.” 

This is the wobbly state of AI-enabled design now, but it comes with a caveat: Whatever it can and cannot do is a temporary condition. Today AI struggles to reason, weigh alternatives, and capture nuance—things human designers constantly do as part of their job. It’s still a blunt and powerful collector, a predictive machine producing stiff but serviceable prose and surprisingly tuneful, if hollow, music, much of it improving in real time. (Its firmer grasp on music is largely because rhythm and notes are predictive mathematical components.) But AI doesn’t yet think spatially; it turns images into two-dimensional pixels. In truth, AI doesn’t actually think at all. “At this point in their development, these systems are fairly primitive,” says Jacob Ward, former technology correspondent for NBC News and host of the podcast The Rip Current. “The diffusion [generative models used for image creation] and large-language models [LLMs] are remarkably good at gathering everything that’s been written, talked about, drawn, and photographed, and compressing all of it and then spitting out a new answer based on that. But that’s not what designers do. They take all of that stuff and synthesize it.” 

The Larger Context

Before making additional educated guesses about AI’s impact on the future of architecture, it’s important to frame the larger context. As much as architects would like us to believe that the profession leads the economic and cultural charge, the truth is more complicated than that. Because architecture is capital-intensive, often moves slowly, requires clients and patrons, and usually has to meet some sort of practical need, architects must constantly adjust and react to a number of contingencies, many of them completely beyond their control: the economy, cultural shifts, labor issues, supply chains, volatile clients, flows of capital, emerging technologies. And yet, even with those constantly shifting contexts and constraints, architecture is, somehow, created. 

While architects tinker with the emerging tools (Midjourney, Firefly, Claude, etc.), global investment in the technology continues to explode. This spending involves two things: products and applications (this is AI’s version of the dot-com boom, but exponentially larger) and physical infrastructure. Between the two of them, it’s a mammoth outlay of capital. Much of the recent growth in the somewhat stagnant world economy is currently tied to investments in AI. In fact, a number of economists believe that without this capital stampede—some of it quite irrational, much of it unlikely to return any profit—the U.S. economy would likely be in a recession. 

And yet for some observers, this frenzy feels, to paraphrase the late Alan Greenspan, irrationally exuberant. In the Ringer article “How Catastrophic Is It If the AI Bubble Bursts? An FAQ,” Brian Phillips writes:

The hype around AI insists that it’s a world-transforming technology that will revolutionize every aspect of human society. The reality … is that AI companies are burning through staggering amounts of money (and fossil fuels) with no clear plan to profitability. … Worse yet for the industry, the biggest players are increasingly tied up in time-bomb deals that look disastrous for their futures in any but the rosiest of best-case scenarios.

What Phillips elegantly describes are the makings of a bubble, set to burst. For the sake of the world economy, let’s hope that AI is indeed a planet-alterating phenomenon capable of defying the laws of fiscal gravity. 

As AI hucksters pitch and run, the large tech companies are undertaking a massive buildup of physical infrastructure. “There are currently 5,437 data centers,” says Frank Stasiowski, a leading management consultant for the AEC industry. “The prediction is that by 2030 there will be about 25,000 data centers.” But this buildup is in no way benign; it will come with a steep price, for everyone. Data centers use huge amounts of electricity and water. Some proposed centers are so large that they will require their own power sources. (Google has plans to build three nuclear reactors.) Most will tap into already stressed local grids and force communities to help foot the bill. According to the New York Times, analysts predict that electric bills around the country are projected to increase by 8% by 2030, and by as much as 25% in states where the AI boom is leading to more data centers being built, such as Ohio, Virginia, and Texas. And make no mistake: these structures are ecological disasters, 21st century smokestacks that pose an even graver threat to planetary survival than the belching towers of the Industrial Revolution. 

So, how much money is riding on this? A staggering amount. Here’s a fun fact: When I asked Google’s AI-powered search engine how much money is being invested in data centers, the bot replied, “Over the next five years, global data center infrastructure investments (excluding IT hardware) could total nearly $7 trillion by 2030.” And while you can’t always trust the accuracy of Google’s search engine (it’s often comically unreliable), I managed to find the same $7-trillion number cited in a McKinsey report. 

It’s logical to think that way too much of this crazy, speculative spending seems untethered from reality. But buried beneath the hype, greed, and giddy excess lies a looming certainty: computers are about to get a lot more powerful, rendering Moore’s Law obsolete, even quaint. “Quantum computing is coming down the line,” Stasiowski says. “Scientists predict that by 2029, all of our laptops will be replaced by quantum computers at 400 to 600 times the speed of today’s computers.” All of these supercharged tools will filter into architecture studios and affect how the work is done. Exactly how remains an open question, but it’s why—despite the noise, pollution, and distinct possibility of a market “correction”—architects must take AI seriously, even if the contours of it are likely to change, and then change again, and again. 

How AI Is Currently Being Used

Here’s a dirty little secret about architects and technology: For all of its glitzy renderings, slick marketing videos, and polished multimedia presentations, the profession has traditionally been a late and reluctant adopter of emerging tech. There are a number of reasons for this that are specific to architecture: cost (new technologies are risky and expensive), liability, and competitive advantage. Architect and author Phil Bernstein, deputy dean at the Yale School of Architecture, cites what he calls the profession’s basic principle of technology adoption: “When you figure out a technological advantage, you keep it to yourself for as long as you can, until somebody else catches up. It’s not as if, for example, there were firms out there helping each other adopt building information modeling.“ 

“[Construction is] even more tech-regressive than architecture, because it can be,” Eric Cesal says. “It doesn’t have the incentives to do deep digital adoption, because such adoption wouldn’t enhance their margins the way that it would in a law firm, or a car factory.”

According to author and futurist Eric Cesal, architects are also slow to innovate because they exist in a supply chain that’s a step directly behind construction, which is full of activities that can’t be sped up by technology. “It’s even more tech-regressive than architecture, because it can be,” Cesal says.“Construction doesn’t have the incentives to do deep digital adoption, because such adoption wouldn’t enhance their margins the way that it would in a law firm, or a car factory. Consequently, architects don’t have much incentive to innovate, either, because if they did, they would just end up creating a backlog, finishing drawing sets faster than the local contractors can execute them.”

So, despite the immense hype, a similar trepidation is occurring now with AI. According to Pratt News’ Alison C. Meier:

A study released this year by the American Institute of Architects (AIA) … found that while just 6 percent of architects surveyed are regularly using AI tools in their work (including chatbots, image generators, and tools for analyzing grammar and text), 53 percent have experimented with AI, suggesting a modest but growing level of adoption as questions about its usage remain.

In other words, they’re using it in the same tentative, exploratory, and often clumsy way everyone else is. But since this is a 21st century technology phenomenon, there is a distinct corporate and generational divide at work. “My sense is that AI is being used tepidly at the firm level,” says Cesal. “I would suspect that a lot of the use is covert. I think you’ve got people in their 20s who are using it much more aggressively than they will let on to their superiors, because the firms are paranoid about liability.” 

In truth, many firms are using AI for everything but design. “Everybody is screwing around with the diffusion models and large language models, the two things that are widely available, creating marketing materials, business plans, generating renderings,” Bernstein says. “The question is: Where’s the innovation?” 

Bernstein is in many ways the perfect source for measured skepticism here. The author of several books, including Machine Learning: Architecture in the Era of Artificial Intelligence, he spent the early part of his career working for César Pelli’s firm, where he managed a number of large projects; served for years as a vice president at Autodesk; and is now deeply involved in the education of tomorrow’s architects, all of whom will use AI in unforeseen ways. And though he’s very much a tech realist and a healthy skeptic, he feels the ground shifting beneath him, too. Bernstein talks about watching a video with a chart showing how long it took different technologies to reach 100 million users: the telephone, Facebook, and DeepSeek. “It was 100 years for the phone, four years for Facebook, and two months for DeepSeek,” he says. “Things are moving quickly, almost too quickly.” 

The question lurking behind all of the AI talk is a simple one: If the technology continues to improve in exponential leaps and bounds, where does that eventually leave the rest of us? Human architects and designers? “I don’t know that it’s inevitable that machines could design an entire building,” Bernstein insists. “There are many places where AI is useful. Where it begins to collapse is when it’s operating in a multivalent environment, trying to integrate multiple streams of both data and logic.” In other words: essentially any architectural project (even a bathroom remodel). 

“Certain streams may become more optimized,” he continues. “If I were a structural engineer, for example, I’d be worried, because structural engineering has clear, robust means of representation, clear rules of measurement. But these diffusion models right now can’t draw a damn floor plan with any degree of coherence.” He describes floor plans as “an abstraction of a much more complicated phenomenon” and thinks it will be some time before these systems can execute the most important things that architects do: make judgments, weigh tradeoffs, exercise experience, take responsibility for what they do.

There’s another limitation as well: the gap between the aggregated nature of current AI data and the disaggregated nature of architectural data. “Maybe, eventually, someone will build something that’s sophisticated enough, multimodal enough, that can operate with language, video, three-dimensional reasoning, analytical tools, cost estimates, all the things that architects need,” says Bernstein. “But it won’t happen in the foreseeable future, unless somebody comes up with a way to train these things on much skinnier data sets.” 

Architectural data, he says, is spread out all over the place, and because it involves risk, no one is keen to share it: “When the Yale Medical School has 33,000 patients enrolled in a trial, they’re getting lots of highly curated, accurate data that they can use to train their AIs. Where’s our accurate data? I can take every Revit model that Skidmore, Owings & Merrill has ever produced in the history of their firm, and it’s not nearly enough data to train an AI. Not nearly enough.”

All of Bernstein’s skepticism feels sharp, on point, and somehow vaguely reassuring (the robots aren’t taking over anytime soon!). And yet, about a week and a half after I posted an interview with him, wherein he complained about AI’s inept floorplans, I received an email from another ardent tech-follower, with breaking news: “I tell people every day, ‘It doesn’t pay to be a Doomer and bet against AI.’ There is now an AI-generated CAD model (editable in AutoCAD) that can accurately draw floor plans. It’s becoming contextual, increasingly able to interact with the real world (built environment). ‘AI can’t draw a damn floor plan?’ Just wait 10 days!” 

A Tentative Glimpse at a Tentative Future

During the course of reporting this essay, Cesal pointed me in the direction of Patrick Chopson, co-founder and chief product officer of Cove Architecture, an Atlanta-based shop billing itself as “the world’s first AI-powered firm.” The obvious question: What exactly does that mean now? Initially, it meant that Chopson and his co-founder, Sandeep Ahuja, were successful in raising more than $36 million in venture capital, a somewhat insane sum for an architectural startup, albeit one with a unique wrinkle. “We began by doing software,” Chopson says, “and then we’ve transitioned in the past couple of years into doing the architecture.”

Chopson makes a point of positioning Cove as “AI-powered but human-centered,” which in practical terms means that the firm’s 10-person software department works in tandem with the architects on staff, applying AI to each part of the design process. In fact, a significant piece of their explorations up to this point have been about analyzing those parts of the process. He calls it the “ontology of design,” adding, “we’re not trying to produce the outcome. We’re designing the process of using AI.” As an example, he takes a five-story garden apartment building. They’ve identified the roughly 87 steps involved to create one, from concept to construction drawings, and mapped out which ones can be executed by computers and which ones must be done by humans. “You always want to keep both things happening at the same time,” Chopson says. “The quantifiable stuff, the details, are what we want the machines to think about, and the unquantifiable stuff—design excellence, beauty—is what we want the humans to think about.” 

Curiously, he’s every bit as skeptical about the current limitations of AI as Bernstein. “We don’t believe that a generalized LLM is capable of thinking about the problem the right way, because it’s based on previous knowledge,” Chopson says. “It’s why only about 10% of our system uses LLMs. The rest of it is code we’ve written ourselves. Obviously, if you have AI generate something based on the most-likely statistical outcomes, then everything will look the same … if you only use AI.” And, perhaps curiouser still, Chopson talks about feeding hand drawings into the model, combining the details handled by the machine with the intangibles often required by architects. “That way,” he says, “you’re not shunted into a particular workflow by an AI image generator that’s trained on the average building of that type.”

“One of the things we’ve learned from the owners is, they’re paying the architect to manage the risk in the coordination of the building,” Patrick Chopson says. “The more you can deliver on derisking their investment, the more they’re willing to pay.”

All of this sounds perfectly plausible and appears to have a logical endgame: significantly reducing the amount of time between a building’s conception and the completion of accurate, finished drawings. Some architects understandably fear that big reductions in that time frame will inevitably result in lower fees and a race to the bottom, once everyone has the same AI tools, but Chopson, as you might expect, remains a tech-optimist. “One of the things we’ve learned from the owners is, they’re paying the architect to manage the risk in the coordination of the building,” he says. “The more you can deliver on derisking their investment, the more they’re willing to pay.” Currently, owners often hire a squadron of consultants to test and validate all of the assumptions made in drawings, essentially acting as backup, tasked with avoiding all of the costly complications of building construction. 

“Speed is also an essential function of how much they’re willing to pay,” Chopson adds. “They’re often taking on big construction loans that have high interest rates. The cost of holding the land while the design process is happening is pretty substantial. For a $50 million or $100 million project, it dwarfs the architecture fee. So you can essentially pay for yourself, and more, if you’re able to execute the job a month or two faster.” 

Chopson’s tech-optimism even extends to what an evolving AI might mean for young practitioners. Even in its awkward iterations, AI is exceptionally efficient at accomplishing, in seconds, what entry level architects have long been tasked with (those pesky details, the grunt work). When I ask him if he had any concerns about that, Chopson’s response is part marketing spin, part messianic belief in the cleansing power of technology (a long-held belief by futurists of every era). “To be honest, I would really like to eliminate entry-level grunt work,” he says. “I would rather have people think about beauty and design, and how the building fits together—the things we learned in school. I would rather our job look a lot more like school and a lot less like ‘I’m drawing bathroom details for the one-millionth time.’ I feel like we’re restoring the soul of architecture and bringing back that kind of master apprentice model from the late 1880s. That’s my goal.” It’s certainly an inspiring aspiration, but not an entirely convincing one. In the immediate term, it’s hard to see how the total obliteration of entry-level work, via AI, helps young people at the start of their careers. 

To Infinity and Beyond

It seems inevitable that at some point soon, two critical aspects of architecture will undergo significant change due to AI: how the actual work is done, and then, in turn, how architects get paid to do that work. The challenging financials are not a new discussion; the traditional business model for architects has been under siege for some time now, and that pressure is likely to intensify. Tools that drastically reduce the cost and delivery time of drawings will force a reckoning. “The business model of AEC firms, which is based on selling hours, is terminal under this new AI-driven condition,” Cesal says. “If the value of production drops, it’s naive to think that all of those savings will be plowed back into design.” 

Bernstein, Cesal, and Stasiowski have all long preached the virtues and utter necessity of what they call “value pricing,” encouraging architects to charge fees based on the financial return of good design rather than the time spent making it. The profession has not heeded this advice, but as technology continues to compress the time required to make drawings, AI might force a long-overdue change. And where does all of this ultimately lead? The Utopians envision a world where architects master, rule, and oversee the emerging AI tools, allowing them to greatly expand both their purview and the market for designed buildings. (The dream of Total Design dies hard.) The Dystopians fear that AI will be co-opted by construction companies and real estate developers, as a fully automated design process is overseen by a mere handful of human designers, who stamp the drawings and cut out the architectural consultants entirely. 

Some tangled combinations of both seem possible. 

As for the actual work? In the past, architects might typically present a fully realized schematic and then enter into a discussion with the client about the pros and cons of each aspect. In five years’ time, AI powered firms might present 10 viable options along with precise schedule and budget calculations for each one. “The actual job is going to be quite different,” predicts Cesal, who envisions a much more managerial role, with architects overseeing the process as a sort of art director guiding multiple intelligences. We are, however, nowhere near that now. “Can AI design a building in its current iteration?” Cesal asks. “No. I don’t think it can replace what an architect can do, especially good architects. But the AI in five or six years, if trends hold, will literally be a million times more powerful than what we have today. Could an AI a million times more powerful do so? I think we’re going to find out.”  

Featured image created by Chatgpt. No AI tools were used in the composition of this work. However, interviews were transcribed by Notto, an AI transcription tool. The completed transcripts required review and revision by the author to ensure accuracy and clarity.