Graduate Studio Examines AI in Architecture: Just Because It Can, Does That Mean It Should?

Supported by UB’s Innovative Instruction Technology Grant (IITG), the studio was a follow-up to last year’s AI literacy seminar led by Anahita Khodadadi, assistant professor in the Department of Architecture. The grant enabled the studio to use the robust FORMAS.AI platform, which was created specifically for designers, rather than more generic alternatives or free platforms that lack the infrastructure needed to appropriately support the studio’s goals—specifically, an in-depth exploration of how AI can reshape architectural workflows. The focus was not on productivity or output alone, but on process and balance: leveraging computational power while preserving critical thinking, authorship, and design intent.

The semester began with an exercise: students input aerial views and 3D models extracted from Google Maps of Buffalo, NY into FORMAS.AI and prompted the system to imagine what the city might look like in 2050 or 2100. The results quickly revealed a pattern: repeated geodesic domes, high-speed transit systems, and drone delivery networks. Because commercial AI models aggregate information from massive datasets, it tends to produce generalized results rather than responses grounded in a specific place. At the same time, the exercise raised important questions about the technology itself—where its data originates, what biases it may carry, and how those hidden assumptions shape its outputs. “This is what we were calling the future of AI but not particularly a future of the place,” Fernando explained.

Recognizing that disconnect was crucial, which is why Fernando chose Buffalo for that first test; the students knew the city well and could immediately see the gap between generic prediction and a likely reality. From there, they worked to reclaim authorship. Instead of relying solely on map-based imagery, they started feeding the AI more intentional inputs—well-written narratives, hand sketches, or a combination of both—framing the tool within their own conceptual thinking rather than letting AI lead the process.

To close out the testing phase, the students incorporated physical models into the workflows. Three-dimensional models gave them more control over proportions than 2D images and reinforced the idea that AI works best as part of a broad design ecosystem, not as a standalone generator.

Fernando was surprised by what came next. Nearly all of the students admitted that they couldn’t get the AI to produce exactly what they had imagined. “That’s something I thought they were going to prove me wrong on,” Fernando admitted. They acknowledged that the criticality and meaning were missing from the results, and they started to question whether that was something that could be embedded into the process. This caused students to look more intentionally at what they understood “data” to be. Their inputs—sketches, collages, study models, scholarly works, precedents studies, etc.—were all now artifacts that contain specific types of spatial information. The foundations of the design process became even more integral towards training these models on architectural ontologies.  

For their next exercise, Fernando recognized that the students would need to understand more about architecture, not just as a design exercise but as a professional practice—the realities of how work unfolds in an office. The task at hand was exploring three questions:

What does an architect’s current workflow look like? Where in that process can AI interject? Do you want AI to participate in that task?

In other words: just because it can, does that mean it should?

While the students had started to develop concerns about the potential of architects being replaced by AI entirely, these initial fears gradually faded as the semester progressed. “They needed to come to that realization themselves,” Fernando explained. “I let them go through the gauntlet, and it was also like a boot camp for skill building. They learned the ethos of the tools, learned how to use them—and the critiques of them—and then they were able to formulate a position about them.”

“It raises more questions than it answers because of the whole aspect of authorship and whether or not an architect is able to retain their own identity,” Aryan Cacodcar (MArch ‘26) reflected. “But I believe the usage of AI opens up a whole new realm of access in the field of architecture.”

Equity implications also came into focus. They began to see AI as a potential equalizer, particularly for smaller firms or individuals with limited technical resources.  AI tools can support competition entries, assist with complex parametric design, or translate advanced ideas without requiring deep coding expertise. Fernando describes AI as a kind of Rosetta Stone, enabling more people to engage with complex systems without needing to have advanced programming skills.

Their next task focused on site analysis: they trained the AI using detailed, place-based information, then evaluated whether the resulting designs authentically reflected the chosen site. Early attempts fell short. To improve outcomes, they found that they had to guide the AI more deliberately: experimenting with combinations of added regional data, refining prompts, and honing their writing skills to more clearly articulate their values and concepts.

As the semester progressed, the students had learned to distinguish between architectural tasks that benefit from AI and those that demand human insight, creativity, and judgment. 

Some students even took their explorations a step further, designing their own AI-driven tools. One created a client-facing interface for single-family housing that allows real-time feedback with the goal of reducing costly change orders. Another developed a tool that caters to contractors and sustainability goals, enabling users to upload projects and see LEED standards, energy metrics, and performance outputs. “That’s where accessibility comes in,” Randy added. “These emerging platforms allow students to engage in tool creation without requiring formal expertise in software or application development. The ability to rapidly prototype custom solutions has opened new conversations around how AI can support highly specific and varied workflows across design practice.”

In addition, the students built an AI tracking app to monitor energy and water use, reinforcing their belief in sustainability as a core design responsibility. In that vein, they also developed a curiosity about the environmental impacts of this studio. They initiated an environmental audit of their process, logging each instance of AI use into that web application. They determined that within two weeks, the studio used an estimated 820 kWh of energy and 773.9 L of water, data that could be built upon through further study to better understand the usage of resources associated with adding AI to architectural workflows, and to potentially develop strategies for offsetting those effects.

For their final project, the students paired up to produce designs to test with newer forms of construction and fabrication, incorporating robotics, 3D printing, and augmented reality. Each team was required to incorporate AI into their process while also making intentional decisions about when to rely on hands-on methods and their own design decisions. A variety of designs resulted: 

AI as Vibe Coding

The students created clay façade panels designed to collect rainwater through a series of simulated contours. AI was used to create custom python scripts that generated the design for the panels. This design hearkened back to the studio’s previous exploration of gathering environmental data to inform future design solutions to help offset AI’s effect on environmental resources. 

AI as Chemist

This project positioned AI as a consultant, using it to develop recipes for biogenic panels of wood and cork shaving remanent from digital fabrication processes in the Fabrication Lab. The materials were bonded using a natural gel agent. The students worked with Claude AI to generate mix designs that could become biodegradable.

AI as Speculative Future

Students used the Fabrication Lab’s Doosan Robot arm to augment their clay 3D printing workflow by allowing human and machine to work in tandem. They used images of their outputs from the collaborative robotic workflow as the input to generate speculative imagery of the future of construction with these tools.

By the end of the semester, the students felt empowered. “They didn’t even realize how much they had learned until they started getting into debates about it and understood  the implications of these technologies enough to take a stance on the topic,” Fernando shared. That’s an important step in developing a critical and informed relationship with AI that empowers them to shape their role in the architectural profession—rather than the other way around.

Acknowledgements:

The studio would like to thank the following people for their contributions via lectures, debates, and discussions during the term: Anahita Khodadadi (assistant professor, Department of Architecture), Manuel Garza (et al. collaborative), Garrett Herbst (Little), Carlos Bañón (Formas.AI), Yiping Goh (Formas.AI), Rutger Huberts (MVRDV), Daniel Eizo (MARVEL), Vivian Lee (University of Toronto + LAMAS), Michael Tunkey (adjunct instructor, Department of Urban and Regional Planning/CannonDesign), César Cedano (Architectural Resources), Tatjana Crossley (Wentworth Institute of Technology + ArchiTAG), Christopher Romano (assistant professor, Department Architecture/Studio NORTH Architecture), Michael Hoover (Studio NORTH Architecture), Gabriela Zappi (adjunct instructor, Department of Archtiecture), Eddie Lam (Architectural Resources)

Students:

Areej Aldar, Siena Allen, Alireza Borhani, Aryan Cacodcar, Lydia Diboun, Joseph Glatz, Shweta Kakade, Shruti Kunadi, Daniel Palumbo, Ada Rodriguez, Edwin Sanchez, Saurav Shetty