AI Can Draw Any Building. But Can It Imagine One?
About six months ago, my friend Steve Barakatt showed me an image of a building he had created with AI. As an architect, I was genuinely shocked.
Barakatt is a Canadian pianist, composer, and music producer — he wrote “Lullaby,” UNICEF’s official anthem. Without any architectural training, he had used AI to design a “piano café”: a building whose roof mirrored the raised lid of a grand piano.
The image was remarkably convincing, and raised an uncomfortable question: If someone who had never drawn a plan or calculated a structure could produce such a compelling design, would there still be any reason to study architecture in the future?
My answer came through an experiment of my own.
In late 2024, “Music on the River” (“Tianmu Qintai”) — a 246,000-square-meter complex on an artificial island in the Tianmu River, in Hengqin’s Guangdong-Macao In-Depth Cooperation Zone — was completed. The decadelong project was the 100th building of my career. Over the following year, I looked back on my designs and wrote a book, “The Stories Behind 100 Buildings.” Friends who read the manuscript told me it did not feel like an architecture book; it felt like a movie script.
That gave me an idea: why not turn my life in architecture into a film?
I had explored other art forms before — sculpture, painting, and musical theater — because I believe architects should be willing to cross into broader forms of visual expression. But film presented challenges of a different scale. Recreating decades of urban transformation and showing designs becoming real buildings would traditionally have required enormous sets and a huge budget. So I turned to AI.
I assembled a 10-person team of screenwriters and technical specialists to turn my story into a screenplay and translate it into instructions that AI models could understand. My role was to guide the results and ensure each frame matched my vision. In less than two months, for under 200,000 yuan ($29,650), we completed the two-hour film “The Architect with Three Shades of Blue,” with a score by Steve Barakatt — the composer whose own AI experiments had inspired the project. The experience gave me a firsthand understanding of both the power and limitation of the technology: artificial intelligence is extraordinary at execution, but it does not create.
This brings me back to architecture. AI’s technical power in the field is undeniable, and its advance is inevitable. Architectural design usually moves through several stages: concept design, schematic design, design development, construction documents, and construction administration. The later stages — developing designs in detail and preparing technical drawings — are exactly where AI can be most useful.
No single person can hold every detail of a massive tower in their head. In a traditional workflow, a major project may involve dozens of people coordinating at once, and communication between humans is never perfect. A window is placed here, a beam runs through there, the two collide, and the problem is discovered only at the construction site. It happens constantly. AI does not make these kinds of mistakes. When a task follows clear logic and defined rules, machines are simply more reliable than people.
The profession is already moving in this direction. According to a 2025 report by the Royal Institute of British Architects, the proportion of architecture practices using AI increased from 41% in 2024 to 59% in 2025. Based on my own observations, mainly from China and Canada, adoption is still accelerating; at large firms, it may already exceed 80%.
Yet AI’s power has a boundary, and it sits exactly where architecture begins: concept design. AI can search for solutions only among things that already exist. It can analyze countless buildings, recombine previous ideas, and produce something that appears new. But it cannot generate an idea that has never existed before.
The concept for “Music on the River” came from the name of the district itself: Hengqin (literally “horizontal qin”). I imagined a building shaped like a qin, a traditional Chinese musical instrument, stretching across the surface of the Tianmu River. Steve Barakatt’s piano café, meanwhile, came from his own experiences and passions. That first moment of inspiration — the spark that connects unrelated things in a new way — is something AI does not possess.
This is why I believe architecture will be pushed toward two extremes. The technical and repetitive parts of the profession will increasingly be absorbed by AI. What remains will be two ends of the spectrum: the rare ability to conceive truly original ideas, and the newfound ability for almost anyone to express their ideas visually.
On one side, the barrier to becoming an exceptional architect is rising. The old architect’s greatest advantage was the ability to imagine and draw. With AI, you only need to describe an idea, and the machine can produce the image for you. The technical threshold has fallen, but the intellectual threshold — the ability to imagine what no one else has imagined — has risen higher than ever.
The architects who can truly create may become increasingly rare. The process through which an architect develops — years of demanding study, endless practice, repeated failures and corrections, and the gradual accumulation of intuition and experience — is being shortened by AI.
I have already seen hints of this in the classroom. At Tongji University and McGill’s School of Architecture, where I have taught, I have seen students whose previous work was ordinary suddenly submit projects of remarkable quality. Were they using AI? I cannot know, and I would not presume to accuse them. But I do worry about their development. A brilliant final submission may conceal an architect who has not yet developed the ability to think independently.
That is the paradox: in the future, an excellent architect may need only a powerful idea and the ability to direct AI. But because of AI, fewer people may develop the depth of experience from which those ideas emerge.
At the same time, the barrier to architecture is also falling. You no longer need to know how to draw; with enough experience and aesthetic judgment, you can use AI to generate the building you imagine. In that sense, anyone can become an architect.
Of course, most buildings created by ordinary people with AI will never become physical structures. But even architects cannot build everything they design. While working on “Music on the River,” I designed an underwater theater: as the canopy opened, the building would rise from the water; after the performance ended, it would slowly descend beneath the surface while fish swam past the audience. The idea was abandoned because it was too expensive and technically complex, and it remains one of my greatest regrets.
Recently, however, I asked my son to recreate that theater from my drawings inside a video game. The idea that players can explore a virtual space and encounter something that I could never realize in the physical world gives me genuine satisfaction.
Six months ago, my friend Steve Barakatt’s piano café made me wonder whether AI would make architects unnecessary. Now I see the question differently. AI may change how we design, but architecture will always begin with the human desire to imagine a better place that does not yet exist, and to allow others to see it too.
Portrait artist: Wang Zhenhao.
(Header image: CoreDesignKEY/iStock/VCG*)*