Membership of Type allows unlimited access to our online library. Join to support new research and writing on the design of the built environment.
You can read more about membership here.
Already a member? Login to your account to avail of unlimited downloads.
Membership of Type allows unlimited access to our online library.
Join to support new research and writing on the design of the built environment.
You can read more about membership here.
Individual Membership Fees
Individual membership is available on a monthly or annual basis.
The fee is €10/month or €99/year.
Contains Spoilers.
The Brutalist was directed by Brady Corbet and written by Corbet and Mona Fastvold. Both were interested in the subject matter due to the parallels between film-making and architecting, in particular the challenges of aligning artists’ creative vision with the expectations of their patrons [1].
Beginning in 1947, the saga spans decades, telling the immigration experience of László Tóth (Adrien Brody), a Jewish Hungarian-born architect. A holocaust survivor who emigrates to America, Tóth eventually comes to the attention of a wealthy industrialist, Harrison Lee Van Buren (Guy Pearce). Van Buren’s commission for Tóth to design a multi-purpose community building initially seems a salvation. Through Tóth’s obsession and Van Buren’s greed, patronage eventually descends to exploitation.
The making entailed nine years of dedication for Corbet and Fastvold (a gestation equal to many buildings). When initial budgets for €28 million made its realisation impossible in Hollywood, it was filmed in Hungary for an incredibly low budget of $10 million [2]. Production design was even hindered by material shortages from the Ukraine war. The entire 3-and-a-half-hour movie was filmed on a very tight schedule, a mere 33 days of shooting. It has been frequently compared to the film Oppenheimer, which had a budget of $100 million and was filmed in a brisk 57 days.
Throughout the film, a number of storylines explore concepts of intent and narrative. When his cousin’s wife accuses László of improper advances, it changes his fortunes irrevocably. We never see evidence of this advance, like many key interactions in this film it is left open to our speculation. However, years later a distraught László references it, saying the allegations were invented because “they do not want us here,” despairing at his incapability to define the narrative as a Jewish immigrant to America. On numerous other occasions in the film, individuals fabricate stories to reflect an imagined or preferred reality [3].
In the epilogue, we are presented with a similar question of authenticity. László’s niece Zsófia, who left America to become an Israeli citizen, presents a retrospective of his work at the first Venice Architecture Biennale in 1980. In her speech she reveals a significant insight: the architecture of the Van Buren Institute was a reinterpretation of the spaces her uncle experienced in the concentration camps. She claims he based certain spaces on rooms in Buchenwald, transforming them with soaring ceilings.
Tóth watches on, wheelchair-bound and mute, as his niece states “I speak for you now”. It is left ambiguous if Zsófia’s version actually was his design intent [4]. She could be retrospectively applying a narrative to suit her world-view, placing Toth’s Jewish identity and trauma at the forefront of his design philosophy and success [5].
We’re told her uncle allegedly outlined an apolitical architectural philosophy in his memoirs, his designs were: “machines with no superfluous parts… they indicate nothing. They tell nothing. They simply are”. This unsentimental outlook gives the second act of the film its name: The hard core of beauty, and the title and theory are lifted from a Peter Zumthor essay of the same name [6]. This is also consistent with one of Tóth’s monologues about architecture earlier in the film [7].
Zsófia ends with a statement that seems to dismiss the creative process and design philosophy we’ve seen in the previous three and a half hours: “no matter what the others try and sell you, it is the destination, not the journey.”
The application of new interpretations outside of a creator’s control, transpositions of meaning, are commonplace in architectural history [8]. As one example, Brutalism, with its muscular, fortress-like forms, is sometimes today associated with federal dominance, even authoritarianism, or the destructive bluntness of urban renewal [9]. At its origin it was often a hopeful, utopian style with ambition to rebuild and rehouse from the rubble of war. The term brutalism originates from raw concrete, béton brut, not brutality. Some film critics have pondered if the ‘brutalist’ in this story is in fact the sinister Harrison Lee Van Buren, applying another new meaning to a brutalist.
Despite receiving ten Oscar nominations, the film has prompted a negative reaction from some architects and architecture critics [10]. It takes many liberties with architectural history; the inaccuracies have been extensively described elsewhere [11]. Its portrayal of the architect as an uncompromising visionary, unwilling to work for others, is reminiscent of Ayn Rand’s problematic Howard Roarke in The Fountainhead. The film’s sombre, serious tone that has led some to incorrectly believe it is, at least partially, a true story [12]. Tied up with the complexities of artistic authorship is the expectation that a serious film like this has a responsibility to be accurate and realist, lest fiction be mistaken for fact.
Many architects and architectural critics find Laszlo’s buildings as depicted unconvincing, particularly so the Van Buren Institute [13]. It is hard to judge the institute, as filmmakers had to be thrifty in how they shot it. Most scenes, for example, had to decide whether to focus solely on floor or ceiling. Only segments of the building were constructed as large-scale models, the rest replicated by computer generated imagery and implied off-camera [14]. A certain number of real sites were used around Budapest to complete the impression. The architecture of the institute is therefore not one thing, a holistic vision, but several fractured things. This portrayal through fleeting glimpses creates a suspense and mystique worthy of a marauding horror-movie monster. Similarly the more we see, the less captivating it becomes [15].
The lukewarm reception of the film’s architecture is all the more fascinating following revelations about its use of Artificial Intelligence. After controversy around the use of AI in post-production to enhance Brody and Jones’ Hungarian accents, an interview with production designer Judy Becker was unearthed. Becker stated that the film’s architecture consultant, Griffen Frazen, used the AI engine Midjourney to quickly create three Brutalist buildings for the film, at an early stage of development. A sample image provided in the article imitates hand-rendering in graphite or charcoal. Becker went on to explain “Now I will have these digital prints redrawn by an illustrator to create mythical buildings” [16]. Corbet has defended the collaboration and creativity of his team, stating that all renderings ultimately used were hand-drawn by artists. A24, however, released a statement that two digital renderings in the end sequence video were generated by AI [17].
With the fleeting glimpses we see of Tóth’s other buildings, it would hardly be a surprise if generative AI was used, even as just a tool in their creation. The buildings appear clunky and varied, mostly resembling incomplete appropriations of brutalism and international-style buildings. These results would be typical of the nascent abilities of AI image generation during the film’s creation (it has already greatly advanced since). Their uncanny quality is reminiscent of what Neil Leach describes as “machine hallucinations” [18]. Familiar yet unfamiliar, they resemble both everything and nothing.
The Brutalist has generated a very rich debate and numerous interpretations (see articles referenced, the list grows daily). Ultimately the architecture in the film is a vehicle, almost incidental to the telling of the characters’ stories. Corbet was less interested in an exercise of faithfully recreating accurate historical architecture, his main intent with the buildings and spaces shown was to externalise the mind of his sullen protagonist [19]. Considering the time and budget constraints on the production, the selective use of AI could be argued as pragmatic.
In terms of who defines the narrative around this film, it's unlikely that the architecture world’s unease with aspects of the film will have much impact. Its enormous success has allegedly generated a new appreciation for Brutalism outside architectural circles, at a time when its buildings are facing widespread erasure from public and private entities [20].
If the film prompts audiences to visit and value the authentic work of architects in post-war America: Breuer, Gropius, Le Corbusier, Rudolph, Kahn, Saarinen, Goldberg, Pei, Yamasaki, Weese; even if one is sceptical of the journey, the destination will be worth it.
Future Reference is supported by the Arts Council through the Arts Grant Funding Award 2025.
1. Feldberg, Isaac. 2025. The Trauma of Inevitability: Brady Corbet and Mona Fastvold on “The Brutalist". 13 January. Accessed 02 10, 2025. https://www.rogerebert.com/interviews/the-brutalist-interview.
2. O'Falt, Chris. 2024. “'The Brutalist' Director Brady Corbet.” Indiewire's Filmmaker Toolkit Podcast. 20 December
3. Van Buren in particular repeatedly makes proclamations on society that reflect his narrow world view. The local town needs a gymnasium since he used to wrestle, but he won’t fund a swimming pool for them since he can’t swim.
4. Adrien Brody has suggested in interviews that we take Zsofia’s narrative literally, however director Corbet is coy in referring to it, and would rather the audience take their own interpretation: "The thing about a piece of public art, and this goes for architecture and cinema alike, is that no one is necessarily right. No one is necessarily wrong”. Stenzel, Wendy. 2025. Entertainment Weekly, The Brutalist ending explained: Director Brady Corbet reflects on building that transformative epilogue. 25 01. Accessed 01 14, 2025. https://ew.com/the-brutalist-ending-explained-8780080.
5. Asch, Mark. 2024. The Art Newspaper, The Brutalist asks who owns the memory of the Holocaust and who defines an artist’s legacy. 30 09. Accessed 02 14, 2025. www.theartnewspaper.com/2024/09/30/the-brutalist-new-york-film-festival-adrien-brody-brady-corbet-architecture-holocaust.
6. Zumthor states: “The reality of architecture is the concrete body in which forms, volumes, and spaces come into being. There are no ideas except in things.” Zumthor, Peter. 1998. “The hard core of beauty.” In Thinking Architecture, by Peter Zumthor, 27-35. Basel: Birkhauser.
7. He states his European buildings would stand outside the politics of the day, perhaps generating different meaning for future generations, stating “my buildings were built to endure”.
8. Whyte, William. 2006. “How Do Buildings Mean? Some Issues of Interpretation in the History of Architecture.” History and Theory 45 (2): 153-77. Accessed 02 14, 2025. http://www.jstor.org/stable/3874104.
9. Campagna, Barbara A. 2020. “Redefining Brutalism.” APT Bulletin: The Journal of Preservation Technology 51 (1): 25-36.
10. Wainwright, Oliver. 2025. The Guardian, Backlash builds: why the architecture world hates The Brutalist. 29 01. Accessed 02 10, 2025. https://www.theguardian.com/film/2025/jan/29/architecture-the-brutalist-marcel-breuer.
11. See Eva Díaz particularly scathing review: Díaz, Eva. 2025. Art Review, The Neoliberal Fantasy of ‘The Brutalist’. 31 01. Accessed 02 14, 2025. https://artreview.com/the-neoliberal-fantasy-of-the-brutalist-brady-corbet-opinion-eva-diaz/.
These inaccuracies persist despite the fact that Corbet and Fastvold were originally inspired by the late historian Jean-Louis Cohen’s book on the architecture of World War Two: Architecture in Uniform, and Cohen was consulted on the film. Schwartz, Alexandra. 2024. The New Yorker, Brady Corbet's Outsider American Epic. 13 12. Accessed 02 10, 2025. https://www.newyorker.com/magazine/2024/12/23/brady-corbet-profile.
12. Notwithstanding some clear parallels between Toth’s life-story and that of Marcel Breuer, the writers have explained their decision to invent a persona: an actual biopic would be open to correction, a fictional one can tell a story unfettered. Feldberg, 2025.
13. Brutalism is often applied liberally to any exposed concrete structures from the 1950s onward but the term was only coined in Post-war Britain in 1953 and stayed largely in the UK until the late 1950s. In a further mélange of architectural language, Judy Becker, production designer on the film, has been open about one of the key inspirations for the Van Buren institute: Tadao Ando’s (much more recent) Church of Light, from 1999. Rao, Anjulie. 2024. Dwell Magazine, How the “The Brutalist” Production Designer Went “Method” to Embody a Fictional Architect. 18 12. Accessed 02 10, 2025. https://www.dwell.com/article/the-brutalist-interview-production-designer-judy-becker-architect-6a15652b.
14. Corbet has described how Brutalism as a style was well-suited to this disjointed approach, more so than an intricate or ornamental architectural style would have been, factoring in the expense and time that would be needed to create a Gaudi project. Despite architects’ qualms with the final built representation, Corbet personally has spoken about relaying indisputable characteristics of Brutalism: the ability to be minimalist and maximalist at once, the ‘minerality’ of concrete. O'Falt, 2024.
15. John Grindrod (who admires the film) describes the community building as “about as unappetising a prospect as you could wish for, a huge blank box with none of the expression of internal functions, asymmetry or sculptural drama and texture that makes brutalism such a beloved – or contentious – style” Grindrod, John. 2025. Recessed Space, The Brutalist: constructing the life of an architect. 07 February. Accessed 02 10, 2025. https://recessed.space/00250-John-Grindrod-on-The-Brutalist.
16. Macaulay, Scott. 15. Filmaker Magazine, Artistic Outputs: Filmmakers and Production Designers on Using Generative AI. 2022 December. Accessed 02 17, 2025. https://filmmakermagazine.com/117846-midjourney-generative-ai/?ueid=f6a2ce9cbbfc3ef017d5e562b8b06ded&utm_source=Sailthru&utm_medium=email&utm_campaign=Gold%20Rush%20-%20January%2024%2C%202025%20-%20TEST&utm_term=Gold%20Rush%20-%20Test%20List.
17. Jones, Nate. 2025. Vulture, The Ghost in the Machinations. 25 01. Accessed 02 10, 2025. https://www.vulture.com/article/did-the-brutalist-use-ai-will-it-affect-2025-oscar-race.html.
18. Leach, Neil. 2021. Architecture in the Age of Artificial Intelligence:. New York: Bloomsbury Visual Arts.
19. O'Falt, 2024.
20. Khomami, Nadia. 2025. The Guardian, Experts hope The Brutalist will revive interest in UK’s modernist buildings. 17 01. Accessed 02 14, 2025. https://www.theguardian.com/artanddesign/2025/jan/17/the-brutalist-brady-corbet-uk-architecture-modernist-buildings.
On the Research & Delivery Team at the GDI Hub, Rebecca delivers applied research to inform more inclusive policies and systems through participatory, user-centred approaches, often with partners in low- and middle-income countries. What follows, after an introduction by the interviewer, is their conversation, lightly edited. Rebecca speaks in a personal capacity; the views expressed are her own and do not necessarily represent those of GDI Hub.
This interview explores research through the lens of disability inclusion research, which is contending with the same questions broader architectural research is: how will AI impact the co-design, to thematic analysis, to the transcripts and workshops that research like this depends on. Building on these shared concerns, this field deliberately focuses on inclusion, a topic frequently underexamined in architectural research.
These issues matter, because our research often shapes the policies and standards our buildings are held to. And we are already slow to design past the minimum of Part M, with research finding only a small minority of new housing in Ireland even functionally accessible [1]. With AI, we can now produce work faster than ever with tools whose biases we may not have critically examined. Their images and renders quietly omit persons with disabilities, allowing exclusion to be embedded in what we make and the possibilities we present to clients, without us even realising. On the flip side, all is not lost, as, when used well, the same tools could open up opportunities for people previously shut out of design, as both clients and as architects. Setting my views aside, I unpick this with Rebecca in conversation below.
AM: Before we get to AI, could you briefly describe your role, and the barriers you're most concerned with in disability inclusion, especially where they meet the built environment?
RJ: I'm a researcher in disability inclusion and assistive technology at the Global Disability Innovation Hub, working on international research and consultancy with a wide range of partners (development banks, governments, disability-led organisations) around the world, with a lot of work in low- and middle-income countries. Much of our research focuses on assistive technology, from wheelchairs and prosthetics to hearing aids and screen readers, through our flagship UK International Development funded AT2030 programme, which tests what works to improve access. The social and human-rights models of disability show how environments, attitudes and policies can be just as disabling as someone's impairment. The built environment is central to that: if a wheelchair user is trying to enter a building that has only stairs and no ramp, they can't yield the benefit of the space, whether that building is their school or their workplace. So we're always looking at the why behind a technology: enabling people with disabilities to live the lives they want and thrive in their communities. On AI, the challenge right now is balance. As a London-based organisation working in partnership with disability-led organisations across the Global South, there are both the opportunities and risks, as well as power dynamics to be mindful of: AI-generated inputs that aren't someone's own words can dilute their lived experience, but AI can also lower barriers, like plain language or translation, for partners and researchers with disabilities. AI should never replace that lived-experience input; at the same time, as these tools spread, I believe we have a responsibility to help level the playing field, including providing guidance on safe, responsible use.
AM: There's a risk in inclusive work of flattening real difference into an 'average' user who doesn't exist. When people come to you through grant applications or co-design, do you think AI advantages those who have it and homogenises everyone else?
RJ: I'd agree. There are opportunities and risks on both ends, and I've definitely seen homogenisation. For research with grassroots organisations, promoting that authentic voice in their own words is really important; when survey responses come back as AI text, there's a huge risk of missing those authentic experiences. AI-generated text should not replace participants voices; outputs have to reflect the thinking, expertise and values of the people involved. So we need to be very clear: for one project, I said we'd rather have voice recordings or bullet notes. We're not looking for polished prose; we really want it to be your voice. At the same time, in a system where professionalisation standards sit within a Western, NGO-ized framework, organisations of people with disabilities in low- and middle-income countries can be excluded from funding for not writing in a certain way, or not having a technical grant-writer. AI can genuinely lower those barriers too. You have to hold both.
AM: Architects and designers lean heavily on renders and images, and researchers likewise rely on images to communicate their findings. As AI image tools often leave disability out, how do we keep our dissemination representing persons with disabilities authentically?
RJ: The AI-generated images are still something I struggle with, honestly. I think it's really important to keep supporting illustrators and graphic designers, including illustrators with disabilities, and to represent that diversity in our work. There's potential for AI to visualise information in new ways that make it more accessible, but surface-level visual appeal should never stand in for depth or authenticity. From my experience, when AI images are used, in a workshop say, it s always declared, and it's usually just a starting point. Sometimes it is to show what something like a journey map might look like; sometimes it is a mock-up a researcher without illustration skills can use to convey an idea, which an actual designer then takes on later. I haven't found it consistent or impressive yet, and that's partly how it works, going off probabilities. We need clear frameworks to uphold responsible, intentional and ethical use.
AM: As a qualitative researcher, where do you draw the line between AI as a help and AI doing the work?
RJ: AI models can act as a mirror of society. There's that term WEIRD: Western, Educated, Industrialised, Rich, Democratic. And we're still a very medical-model society, viewing disability as something to be fixed, with people with disabilities often excluded from the datasets these models are trained on. Those same biases get baked in, so you have to check for them. With thematic analysis, AI tools can help summarise or paraphrase to support initial coding, but that should still be human-led, with human review. There s also ongoing debate about whether AI should be used for this at all. Braun and Clarke, who are the go-to authors on reflexive thematic analysis, even put out a paper arguing against using generative AI. I think of a colleague, a public-health researcher and data scientist, whose expertise on a project in India was essential to validating how we framed our statistical claims. That kind of judgement is exactly what I don't think AI can replace. I believe we should use AI in ways that enhance our value, strengthened through critically interrogation and human expertise. And wherever AI meaningfully contributes, it should be documented and disclosed, for transparency and research integrity. One more thing worth flagging is data protection. Using these tools often means handling personal and special-category data, the sensitive kind, including health and disability information. Where that data goes, and how it is stored and used, needs to be a serious consideration every time, and even more so in work with disabled communities.
AM: When AI is used to carry out a literature review, do you think there's a loss in the knowledge a researcher would otherwise gain from doing it themselves?
RJ: I've used tools like NotebookLM and Research Rabbit to organise sources and draw connections, but if you let the AI choose your sources, its search strategy has its own biases, and there's real knowledge lost. If you're not actually reading the literature, you're not gaining what comes from doing the review.
AM: On the more hopeful side, could AI widen who gets to take part, and who gets to be a researcher?
RJ: There's a lot of potential there, for example, with accessible science communication and access to knowledge. We have Founders Think at GDI Hub, built on our accelerator knowledge, so a designer or entrepreneur with a disability who faces barriers to entry to an accelerator could work with it on their own time. People with disabilities often face reduced social participation and smaller networks, and digital barriers can shut them out further, even from applying, if a website isn't accessible. Not gatekeeping that knowledge is a real opportunity. It also has applications in inclusive education, like simplifying language to different reading levels, which can help someone with low literacy or a learning disability. So there's genuine potential for accessibility, alongside the downsides around biased data.
AM: This is something I feel strongly about. If the people trying to do good step away from AI, we leave it to be shaped by everyone else. Do you think we have to engage with it rather than opt out?
RJ: So many people are trying to use it for harm that we have to be in there too; it learns from us. I think about the Centre for Digital Language Inclusion, a GDI Hub project using AI to support people with non-standard speech in African languages, a brilliant example of AI for good. But at the same time, more broadly, there's the risk that generic AI transcription tools don't capture those same people. It's often two sides of the coin. As a paper by Musslick and colleagues put it, the use of AI in science is not a neutral act: it reshapes workflows, influences which questions get asked and affects how evidence is weighed and interpreted .[2] So in my view, we need to upskill rather than turn away, keeping the expertise that lets us stay nuanced and staying transparent about how we use it.
AM: If we spoke again in ten years, what would tell you AI had been a positive influence on your field?
RJ: I'd love to see it democratising access for people with disabilities to become researchers and reach scientific knowledge, reducing those barriers to entry. That would be a really amazing outcome. And I'd want a clear framework on best practice for AI in disability research. The technology has moved too fast for us to catch up; without a shared framework it's hard to know what, when and how to apply it ethically. There are general principles for responsible AI, but there's a uniqueness to who we work with: people with disabilities in low- and middle-income countries at the very edge of exclusion. So we need something specific to that. Right now it seems like a bit of an impossible dream, but I'll put it out there anyway.
What stayed with me from this conversation was the importance of engaging with AI rather than stepping away from it. Rebecca and I share concerns about the biases these tools can reinforce, the voices they risk excluding and the knowledge we lose when we allow them to do our thinking for us. Yet we also recognise their potential to open doors for people historically excluded from research and design. Rebecca also reminded me of the low- and middle-income contexts I have previously worked in, and where she continues to work, where people with disabilities already face disproportionate barriers to everyday life, including access to basic needs such as clean water and reliable energy. In this context, the environmental cost of AI is not simply a concern for climate specialists, such as Erida who I spoke to last month, but a question of equality and who bears the consequences of technological progress. Perhaps the challenge is not simply how we use AI to make our work more inclusive, but how we ensure that, in doing so, we don't create new forms of exclusion.
AI declaration
This interview was conducted in conversation and transcribed with AI-assisted transcription using Microsoft Teams; the transcript was then edited by the author for length and clarity, with the interviewee's approval of the final text. The accompanying image was generated using GPT-5.6 Sol, guided by the author's careful prompting to represent the themes of the discussion abstractly rather than to depict any real person, place, or event.
In the second issue of our mini-series 'the mainframe', Annamae Muldowney speaks with Rebecca Joskow, a researcher and practitioner in disability inclusion and assistive technology at the Global Disability Innovation Hub (GDI Hub), a research and practice centre driving disability innovation for a fairer world, based at University College London (UCL). They discuss AI, inclusion and the built environment.
Read
As a consultant at Buro Happold, Erida Bendo advises clients on climate-responsive design at building and urban scales, drawing on microclimate modelling, environmental analysis and spatial data. Alongside this, she leads computational design for the firm's Sustainability team, building the tools and workflows behind evidence-based decisions. She holds a Master's in Architecture from the Polytechnic University of Tirana and a Master's in Computational Design from IAAC in Barcelona, where she has also taught. Over the better part of an hour we covered speed, scepticism, and why the most important climate knowledge still sits between the disciplines. What follows is our conversation, lightly edited.
AM: Before we get to AI, how would you describe your role, and the main climate challenges you're dealing with?
EB: I can describe that best by talking about the two scales the projects I'm involved in revolve around. There's the urban scale, where I mostly consult on climatic aspects of masterplan designs. The thought process begins with understanding which climatic qualities we would like to enforce on outdoor spaces. How to design places that are less prone to heat stress, which support natural ventilation without creating wind hazards during winter is a recurring theme in that sense. So it's mostly about designing with the sun and designing with the wind.
And the same principles are applicable when we talk about the building scale, where I look at the building envelope: how to shape geometries and orientations, what materials are advantageous, how to design the openings in a way that creates good daylight but less overheating? Programme considerations and occupancy, for example what functions these spaces are hosting, during which times of the day or of the year, can also be other decisive factors, which can lower the reliance of buildings on mechanical systems. My interest here is mostly around the effect of passive measures. I use simulations for reassurance, but I wouldn't say they're currently the main bulk of my work, or what I consider the challenge in my work.
AM: So where do you see AI having the biggest impact on your area?
EB: There are two promises of AI: it's supposed to make the work better, faster. On ‘better’ I'm not very sure, but I'm more or less convinced by now that it does make these processes faster. The catch is that it also closes the discussion early. Several years ago, an architect couldn't run a climate simulation themselves, so they'd have to bring specialists in and that meant a real back-and-forth, several of us working on the problem together over time. Now, with all these tools, increasingly AI-assisted, the architect can run a simulation, get a result, feel they've understood the question, and the discussion just stops there. They work faster, yes, but the conversation ends early in the day, before many people have had a chance to shape it together. What I'd expect in the future is fewer people involved in each project, and the design process finished faster.
AM: Could AI make climate and sustainability expertise more accessible to architects without specialist knowledge? And is there a risk it creates an illusion of expertise i.e. that it flattens or dilutes what a specialist really knows?
EB: I definitely agree. I'd compare an AI assistant or any other algorithmic workflow to a human specialist. You have professionals who are very well versed in what they do, and others that are not or who are maybe still figuring things out. So there will be cases where AI supports you and makes things go better and faster, and cases where it creates misunderstanding, specifically that risk of the illusion that you've actually fulfilled the task when you haven't. It's the same as having a specialist in the loop: it depends on the knowledge and the quality of the workflow you're using. And the startup scene is so alive right now, every day there's a new software that can run all these simulations. So with what's out there currently, I do see that effect of creating the illusion. Maybe in the future it won't be the case, but I'd definitely say it's quite a risk.
AM: Is there an area you'd not want to see AI take over?
EB: I can see AI being applied in very specific use cases, but I wouldn't like to see it as something that guides the design process. I'm not a fan of configurators, for example: workflows where you can generate masterplan or layout designs with a mouse click. It goes back to the distinction between holistic technologies and prescriptive technologies, which Ursula Franklin has noted in her work. There's this distinction, prescriptive is when you know every step you need to achieve; and then you have holistic technologies, also linked to crafts, for example pottery makers. And for me, architecture is a craft. So I wouldn't want a design workflow where you press some buttons and you get a finished building out of it, because design as a process isn't only about achieving certain KPIs. It's also about involving communities, involving people whose lives actually get affected by those decisions.
AM: Then where does human insight hold its value?
EB: The role of the climate specialist specifically is changing. If you look at the last century, it was doctors and not architects who were keen on designing ventilation schemes for buildings because of health hazards. With the advent of air conditioning mechanical engineers came into the loop. Then, some years ago, and I'm part of this group, architects with computational design skills started emerging, who got hired in positions where they would carry out building simulations. And now we're at the point where that might not be needed at that scale, because the threshold for using computational tools has been lowered. So I think it's about having the knowledge that sits at the intersection of different topics. It's not only the climate topic, if you look at constructability, or material topics, the building industry is also a very local thing, it differs from country to country. Having the knowledge that sits at the intersection of all of these, and consulting on the best alternative in terms of climate for a given site, given people, that's something I see myself moving towards. And again, it's not a given that simulations always truly reflect what’s really happening out there, the real world physical phenomena. So there's also a process of learning from past projects, from the way you live in your city, the buildings you visit. It's not a discussion that can be merely reduced into a very prescriptive way of following the same steps.
AM: Considering insights, can you ever really hold an AI accountable for what it tells you?
EB: That's the point I'd raise also: can you actually hold it accountable for what it's saying? It takes time to develop trust when working with other professionals, you learn who they are through hearing them speak about who what they do, notice how versed they are on topics, observe how they live their lives and what beliefs they share. I feel that this does not happen while interacting with AI-generated answers. I remain sceptical.
AM: Where would you most want to see AI help on climate specifically?
EB: I'd see it in trying to better understand physical phenomena. What most of the time happens with these climate AI-assisted simulations is that they're built based on results of simulations, they're not built based on what's happening out there. It can be something build on top of simple machine-learning models as well. It's difficult, of course, because you have to get out of your office and try to build a good database. That's where I'd actually be interested in seeing more work done, something that extends the capabilities of simulations, because they're very idealised situations where a lot of approximations happen
AM: As you have mentioned environmental decisions rest on assumptions about a future climate we can't really predict. How do you design for that uncertainty rather than for a single scenario and is AI any help there?
EB: The current discourse revolves around climate models. A lot of climate research centres have simulated different scenarios. But even within that research it's not necessarily clear what we should expect for the same scenario, if you open the datasets from different research centres, you get very different results. So it's not about trying to find what exactly will change, because we won't be able to know that. It's about trying to make more robust buildings. The key word there is robust: don't prepare for a specific future, but for a range of them. As for how AI fits into that, maybe it can help in predicting better the effects of climate change, honestly, I don't really know; for me it's more about changing our approach.
AM: AI has its own environmental footprint. How should the industry hold that tension?
EB: This is not a problem that comes from the building industry itself. In academic discourses architects always put themselves in this position where they have to improve the future, but actually it's a problem about how we live and how we've structured our economies. We're not building because we have skilled architects; we're building because there is an economic profit in doing so. So it's not just about pointing to our individual decisions, pointing to every individual about what they can do best. It's about treating the system as a whole.
AM: If we spoke in ten years, what would tell you AI had a positive influence?
EB: I'd be happy to be in a position where AI takes care of the boring small steps along the way: for example supporting interoperability between different software, different specialities - if I never had to take care of 3D models again. It's this idea of using it for the less interesting bits in the design. And I'd find it dreadful, actually, to see configurators where architects are just people who press buttons.
AM: So would AI have a positive influence if it brought more people into those multidisciplinary, cross speciality conversations, helping everyone hold their own place at the table, and making sure each voice is actually heard?
EB: Definitely. Now that you say it like that, it could also be a kind of project manager, a position that actually makes sure people are heard, since because of many different factors you do have teams that speak about certain aspects more than others. And what I mentioned earlier, the data discussion, just ensuring a building model can be used between different specialists remains to be a challenge. Perhaps AI as a technology could support the emergence of more intuitive tools that support a holistic design process, but then again for me the decisive factor in this discussion remains the professional, if they choose to be aided by algorithmic processes, a level of technological skills would need to exist so that they do not diminish to being just a prompter.
What stayed with me from this conversation is how closely Erida's concerns mirror the ones I meet in my own architectural speciality. I came in expecting the challenges of environmental design to be a world away from inclusive design, and instead we kept landing in the same place: the danger of averaging, of optimising for a single predicted case, be it a person, or a climate, that doesn't really exist, and of mistaking a clean simulation result for the messy reality it stands in for. Her answer, again and again, was to keep people in the loop, the communities a building serves, the specialists around the table, the human judgement that reassures rather than the tool that decides. It's a quietly hopeful position. Not that AI will make us better designers, but that, used with some scepticism and for the right things, it might free us to spend more of our attention on the judgement, the context, the craft.
AI declaration
This interview was conducted in conversation and transcribed with AI-assisted transcription using Microsoft Teams; the transcript was then minimally edited by the author for length and clarity, with the interviewee's approval of the final text. The accompanying image was generated using GPT-5.6 Sol, guided by the author's careful prompting to represent the themes of the discussion abstractly rather than to depict any real person, place, or event.
In the first issue of our new mini-series ‘the mainframe’, Annamae Muldowney interviews Erida Bendo, a Berlin-based architect, computational designer and environmental design specialist. Through expert interviews, the mainframe examines the impact of Artificial Intelligence on design. Each interview focuses on a different intersection, with this issue focusing on what AI can, and cannot, do for climate.
Read
These difficult questions are not idle speculation. The capabilities of AI are increasing by the day, and our long-held convictions on creativity and design are being questioned [1]. Personally, I have transitioned from working fifteen years as an architect and I now lead AI development, strategy and research at a large architectural practice. I have been observing these tools being used at every project stage and can see areas where they are working and are not. One thing I believe is certain, is that the way we have worked previously is now broken.
Irrespective of whether you're sceptical on AI, unconvinced by what you've seen, or if you've already integrated AI tools in your armoury and are familiar with how they are transforming work from the inside, the context to AI's role in creative work is constantly and rapidly changing. Some are less concerned about the capabilities of AI and more about the consequences for the industry, the values, and above all, its impact on people. These are all valid concerns.
Whether AI can design, and whether AI can be creative, are two different questions. Conflating these questions is where most of the current debate loses its footing. AI's capacity to design, in the sense of performing the tasks that constitute a design process, is largely a question of model capability, and the answer is changing at a pace that is difficult to keep up with. We are arriving at a point where AI agents can begin to orchestrate parts of the process, but without meaningful guidance they have no understanding of why they are doing what they are doing. The creative process is not always linear and is often not compatible with delegation. It does not move through predictable stages with clear milestones. It is continuous, unstructured, at times chaotic, and the understanding that guides it is often something a designer knows intuitively but can often find difficult to articulate, even to themselves. An agent can follow a sequence, but it cannot feel its way through one.
Whether AI can be creative is a question of a different order entirely, one that sits closer to what it means to understand something, to care about it, and to make something in response to that understanding. Creatives are perhaps better placed than anyone to navigate this technological shift, because the answer to the article's question has less to do with what the technology can do and more to do with what we can and will always bring to our work.
Architect and theorist Christopher Alexander devoted much of his career to a question that is simple to ask and very difficult to answer: why do certain places feel deeply, immediately right in a way most people sense but few can put into words. Alexander described design as a search for good fit between form and context, where context meant not a background condition but the full weight of human needs, constraints and relationships easy to miss unless fully understood; 'We are searching for some kind of harmony between two intangibles: a form which we have not yet designed and a context which we cannot properly describe'[2]. His concern was that reducing design thinking to a transferable system passes on the logic but loses the life. Production is a large part of the work we do, but the harder challenge has always lain elsewhere. That gap between systematic knowledge and embodied understanding is exactly what AI now forces us to confront again.
Ethan Mollick, a professor and leading researcher on AI & innovation and its impact on society, describes the form of AI we have ended up with as 'deeply weird in ways that we don't fully understand yet'[3], and warns that treating it like any other tool will always produce less useful outcomes than implementations that embrace that weirdness. My observation in architectural practice, is that the people who are most willing to lean into that strangeness are the ones most capable of influencing design direction, approaching these tools out of deep curiosity [4]. AI only flattens creative work when it is used to seek the average and remove judgement from the process. When designers invite the strange instead, it can lead to something genuinely intriguing.
What AI tools can offer, more than anything else, is freedom. Freedom to explore further, to reach into areas that once felt out of range, to test an idea without the weight of technical limitation slowing the thinking down. Designers are following their curiosity into new territory and finding that the boundaries they once worked within were never as fixed as they seemed. The curious are building their own tools entirely, which is perhaps the purest expression of that freedom, moulding the technology around their imagination rather than the other way around.
We come to the realisation that the process can be delegated, but the understanding behind it cannot. This is not a new concept, and it has always been framed as something existential. CAD was going to be the demise of the art of drawing, CGI was going to hollow out cinema, the sewing machine was going to end fashion as a craft and of course the video killed the radio star. Each time, the creative industry absorbed the tool, expanded its reach and moved onto the next challenging question. The pattern is consistent enough to resist the urge to panic.
So, is creativity still inherently human? My immersion into the space between suggests to me that the answer is yes, and the more capable these tools become, the more important it is to understand why. What AI offers is the removal of friction between a designer and the full scope of their thinking, and while that is incredibly valuable, it is not the same thing as being creative. Creativity is not something the tools produce. It is what we bring to them, the direction we set, the judgements we make, the willingness to keep questioning whether the work is right until we believe that it is.
Alexander asked this question before the tools existed and arrived at the same place: creativity lives in understanding, and understanding remains ours to develop or to neglect. The future belongs to those willing to embrace curiosity.
Disclosure of the use of AI is an important aspect of the work that I do. For transparency: I have used Wisprflow to dictate my thoughts, and Anthropic's Claude Sonnet 4.6 to map these themes for the article concept. The article was edited in collaboration with Cormac from TYPE through phone conversations and document exchanges. The images throughout have been generated with MidJourney. The content and ideas behind the article are my own.
Creativity has long been the human capacity we considered beyond the reach of any machine. Most can agree that Artificial Intelligence (AI) has crossed the threshold of being on the periphery to our work and is now embedding itself into our thinking, our workflows, and our society. As these shifts begin influencing the creative industries, we have to ask: what truly changes, and is creativity still what makes us human?
Read
Website by Good as Gold.