AI in the studio: a new production regime

It is undeniable that AI tools are increasingly embedded in architectural practice, from professional offices to educational settings. In my own recent experience, this is perhaps most visible in the design studio—the backbone of architectural design curricula—where generative AI tools have become particularly attractive, as they enable students to rapidly produce polished-looking drawings, images, and narratives, which constitute the primary means through which design intent, logic, and resolution are communicated.

This is not surprising. Studios have always been environments where modes of production matter, and tools that accelerate or enhance representational capacity tend to be experimented with and adopted quickly. What feels different this time is not simply the speed or quality of the outputs, but the way these tools reshape the relationship between process, iteration, and apparent resolution.

Studio pedagogy and modes of production

This shift raises concerns on both pedagogical and ethical fronts. The ethical implications of AI use in architectural education—questions of authorship, attribution, intellectual property, and transparency—are significant and deserve careful consideration. They are not incidental concerns. However, I am intentionally setting those questions aside for now. My focus here is not on regulating AI use, but on understanding what design education is fundamentally trying to teach, and how emerging modes of production might reshape or clarify that task.

From a pedagogical perspective, the ease with which finished-looking material can be produced risks short-circuiting the experiential dimensions of design learning, traditionally associated with drawing and model-making. In most of the studios I’ve been involved with over the past decade, these practices have functioned as the core didactic building blocks of studio education—not merely as representational techniques, but as mechanisms for reflection, iteration, and critical engagement.

Through drawing, making, and revisiting propositions, students are encouraged to think through material, organisational, and spatial problems. Narrative, in turn, emerges as a way of articulating that reflective process: a means of explaining why decisions were made, how ideas evolved, and what trade-offs were negotiated along the way.

Equally important is the way analogue processes tend to unfold from low to high levels of detail. Early sketches and rough models are deliberately ambiguous. That ambiguity is pedagogically productive: it invites reinterpretation, generates alternative readings, and creates space for new ideas to emerge. As projects develop, increasing levels of detail correspond to increasing levels of certainty. The gradual resolution of form mirrors the gradual clarification of intent.

When each stage of a process instead produces outputs that appear finished or resolved, this ambiguity can be prematurely compressed. Generative systems do not eliminate iteration altogether—AI-generated outputs are rarely satisfactory on the first attempt—but they rely on fundamentally different forms of input. Iteration occurs through prompting (which in many cases is multi-modal), selection, and adjustment, and exploratory processes can become structured around the comparison of seemingly complete alternatives rather than the incremental emergence of ideas.

This raises what I see as a key pedagogical question: what are the skills that studio education is fundamentally trying to develop, and how might new modes of production reshape — or clarify — how those skills are taught?

From tools to skills

Given the extent to which these technologies have already permeated daily life, and by extension both education and professional practice, I am sceptical that attempts to simply police or prohibit their use will be productive. At the same time, the exploratory potential of these tools is hard to dismiss.

The more interesting question is not whether AI should be used in design education, but how studio pedagogy can continue to foster critical thinking, synthesis, spatial problem-solving, systems thinking, and analytical rigour while engaging with new modes of design production. This also requires asking which skills must be redefined, supplemented, or newly developed as AI becomes part of the designer’s toolkit.

It is important, at this point, to acknowledge a common concern: analogue techniques offer a particular kind of immediacy. The close coupling between action and perception in drawing or model-making supports continuous adjustment and refinement, and that immediacy has long been central to studio pedagogy. However, the pedagogical value lies not in the medium itself, but in the designer’s capacity to interpret feedback and steer iterative processes toward meaningful outcomes. AI-mediated workflows reconfigure this dynamic rather than eliminating it. Feedback becomes less continuous and more comparative; iteration shifts from gesture to judgment. The challenge, then, is not to preserve specific feedback mechanisms unchanged, but to ensure that students continue to develop a robust capacity to work with feedback, regardless of how it is mediated.

Rather than centring the discussion on specific tools, this shift invites a clearer articulation of the underlying competencies that architectural education aims to develop—competencies that should persist even as modes of production change.

Design skills that translate across modes of production

What follows is a working list of design skills that, in my view, remain central regardless of whether design is developed through sketching, model-making, scripting, 3D modelling or prompting.

Problem framing: The ability to identify what the problem actually is, distinguish causes from symptoms, and translate vague briefs into actionable questions. This skill exists before any drawing is made or any prompt is written.

Contextual grounding: Design operates within regulatory, material, environmental, social, and economic constraints. The capacity to recognise, prioritise, and work productively within these constraints is a core design skill. While AI can generate options, it cannot determine which constraints matter most in a given context, unless those constraints are clearly defined by the designer.

Critical evaluation: Good design depends on the ability to assess alternatives against explicit and implicit criteria, compare alternatives in terms of performance as well as appearance, and make decisions under uncertainty. This capacity becomes more—not less—important as the number of available options increases.

Design reasoning: The capacity to construct and follow a coherent line of inquiry through a project — knowing why a direction is pursued, what is being tested at each stage, and when a line of exploration should be abandoned or refined. It is not about repetition or variation for its own sake, but about maintaining a legible logic that connects decisions across the design process. This applies equally to all forms of production."

Critical synthesis: Architectural expertise builds through the intelligent reuse of precedent, typologies, and strategies. AI systems make this process explicit by operating through recombination, but the skill lies in selecting, adapting, and contextualising existing knowledge rather than reproducing it uncritically.

Design responsibility: Ultimately, designers remain responsible for the consequences of their decisions. This includes the ability to explain why particular choices were made, to acknowledge limitations, and to take responsibility beyond claims of authorship or originality.

Taken together, these skills point toward a different set of criteria for evaluating design work — criteria that are less explicit in studio culture than they perhaps should be.

Studio culture and the problem of proxies

Beyond the question of AI itself, architectural design studios have long been predicated on the principle of “learning by doing.” In my experience, this approach genuinely seeks to engage students in design as an applied process, grounded in challenges related to inhabitation and use, and how these are ultimately translated into physical entities. Yet I have also observed that, in practice, this pedagogy often carries a persistent expectation of innovation, understood primarily as the production of solutions that appear formally and aesthetically unique.

Within this context, there can be a subtle reluctance to openly acknowledge references, precedents, or forms of inspiration, as design proposals are often expected to present themselves as original and purposefully authored for a particular task. In many studios I’ve been part of, creativity becomes closely aligned with formal differentiation, and aesthetic distinctiveness is sometimes treated—both implicitly and explicitly—as a proxy for design quality. As long as producing such distinctiveness requires sustained effort and skill, this proxy appears credible and is rarely scrutinised.

Generative AI tools—whose operation is explicitly grounded in the synthesis and recombination of existing material—make this assumption harder to sustain.

AI as catalyst, not cause

Rather than undermining studio pedagogy, I see this shift as an opportunity to reorient design education toward the practical and evaluative dimensions of architectural work. If formal differentiation and aesthetic variation can be readily manipulated, their role as primary indicators of value becomes questionable.

This foregrounds a more fundamental question—one that is worth discussing openly as researchers and educators: to what extent are the aesthetic aspects of a design project genuinely tied to its architectural value, and how might architectural education more explicitly recognise, teach, and assess the skills that underpin good design, regardless of how that design is produced?