How is AI Impacting Architecture Today? Six Takeaways from a Harvard GSD Virtual Town Hall
Illustration: Ihor Tsyvinskyi
Today’s Harvard Graduate School of Design (GSD) Virtual Town Hall posited that AI will not “replace” architects so much as it will rewire how they learn, decide, and collaborate. Dean Sarah M. Whiting framed the conversation as a pedagogical question, then pressed presenters Martin Bechthold, Humbi Song, and Eric Rodenbeck on what we teach now, what we protect, and what we refuse to outsource.
What the panel made clear about AI’s impact on architecture today
1. AI compresses the path from question to concept, but it does not eliminate judgment.
When Whiting asked the anxious version of the question (“Will architects just speak to machines?”), Bechthold responded with a practical reality: the tools change weekly, but design still requires informed expectations. In his view, AI can reduce barriers to technical work, enable teams to address conceptual problems more quickly, and facilitate collaboration among mixed-skill student teams. It still requires a designer who can assess credibility and evaluate results and outcomes, as we already do through simulation and analysis. In other words, AI accelerates iteration; it does not provide discernment.
AI accelerates iteration; it does not provide discernment.
2. The biggest near-term risk is not “automation.” It’s sameness.
Song named what many designers already sense when they scroll: the “AI architecture” feed trends toward visual clichés. She connected this to a cluster of studio risks: an illusion of knowledge, the erasure of individual perspective, flattened materiality, and “design fixation,” in which a tool returns something plausible too early and a designer stops exploring alternatives. Her core critique targeted the common A-to-B interface of many generative tools: you input a hazy thought and receive an image that looks resolved, whether or not it carries your authorship. That dynamic can narrow design space at the moment when it should stay open.
3. The most promising AI workflows look less like “prompting” and more like “thinking with tools.”
Song pushed the conversation beyond Midjourney-style image output to real-time, iterative creation. Her examples mattered because they kept architecture’s embodied intelligence in the loop: students sketching on a board while a projector overlays AI variations in real time; a student prototyping with scrap cardboard while AI helps her see options without replacing her assembly knowledge; a VR-based design process where generative output responds to what the designer draws at full scale, then returns to the physical world through a carved piece of discarded marble. This approach treats AI as a responsive medium within an iterative craft practice, rather than as a vending machine for finished form.
4. Transparency will become a baseline professional norm, not an academic courtesy.
Rodenbeck offered the simplest rule with the biggest cultural consequence: “You can’t lie about it.” In his course, if you use AI, you disclose it, including prompts. He highlighted a student project that turned that disclosure into the work itself. Roy Zhang repeatedly asked ChatGPT the same question, tracked the shifting keywords, and mapped how the variation eventually converged. The point was not to generate “better” outputs; it was to interrogate the act of asking and to expose the model’s landscape of answers. That posture is directly applicable to practice as well.
5. AI expands what small teams can attempt, which will reshape practice economics.
This theme surfaced in Bechthold’s comments about workflow design and in Rodenbeck’s concrete examples. A student used machine learning to identify typefaces across a large portion of MoMA’s graphic design archive and assembled a “museum of fonts” in about a week. The toolchain was imperfect, and the result later failed, but the capability matters: a student or a small firm can now conduct analysis and synthesis that once required specialized staff, long timelines, or both. That shift will not only change deliverables. It will change who can compete, what “scope” looks like, and how firms price insight versus production.
6. Pedagogy will need one major upgrade: teach students to notice how tools shape thinking.
When Whiting asked what should change in the required curriculum, Song’s answer was poignant. She argued for awareness: tools mediate design decisions, often invisibly, and the first weeks of school create a rare “teachable moment” because the mediation still feels visible. Her second point expanded upon another point directly aligned with professional practice: architects already design workflows. AI simply accelerates that long-standing reality by enabling more people to “buy code” and build custom pipelines faster. The curricular opportunity is to help students become conscious workflow authors who can branch, iterate, and return to earlier steps, rather than treating design as a linear march toward a single output.
The takeaway for architects today
The conversation analyzed the myriad ways architects can use AI to iterate, produce, and analyze. AI will change the pace of architecture, broaden access to certain types of analysis, and intensify pressure on authorship and ethics. At the same time, the panelists kept returning to a central reality: architecture lives in three dimensions, with human bodies, tacit knowledge, and material consequences. Those realities do not reduce to data.
Architecture lives in three dimensions, with human bodies, tacit knowledge, and material consequences. Those realities do not reduce to data.
Whiting summarized the school’s posture: the GSD teaches the ethics of acknowledgment, authorship, boundaries, and collaboration. That is not a defensive stance. It is a strategy for staying human and staying credible in an AI-influenced practice.
AI disclosure: this piece was aided by transcription via Otter.ai, copyediting by Grammarly, and ChatGPT-generated summaries.