Source: Architizer and Chaos, The State of AI in Architecture, 2024 Survey: 1,227 architecture and design professionals across 118 countries
Architizer and Chaos surveyed 1,227 architecture and design professionals across 118 countries about the use, limits and expected role of AI in practice. Most respondents worked in small practices: 63% were employed by firms with twenty people or fewer.
Adoption is real. Training is not.
46% of respondents were already using AI tools or features in architectural projects, with a further 23% planning to use them in the near future. Adoption rose with firm size, from 35% among freelancers to 55% in firms of twenty or more people.
Against that, 60% had received no formal AI training, while a further 18% planned to receive some. Much of the profession's early AI practice is therefore self-taught, improvised and weakly documented.
Whatever best practice eventually exists will be written by today's experimenters.
It is used almost entirely for images.
The dominant applications were image generation from text prompts (74%), image editing (61%) and image generation from model inputs (50%). Other use cases remained much less common: layout and plan generation (26%), feasibility studies (21%), building-code analysis (19%) and energy-efficiency analysis (19%).
Satisfaction followed the same pattern. 67% were satisfied with AI-generated renderings for early design, but only 30% were satisfied with its use in design development and later stages.
Architects are adopting these tools where work remains exploratory and resisting them where output must become coordinated, documented and defensible.
The complaint is control.
The most frequently cited obstacle was limited AI functionality for architecture (50%), followed by integration problems (38%), missing training resources (37%) and lack of time (37%).
When asked what they wanted next, respondents prioritised greater control over output images (62%), conversion of a 2D image into a 3D scene (58%) and material generation from text prompts (55%).
None of this is simply a request for more capability. It is a request for predictability.
Expectation runs ahead of evidence.
86% of respondents expected AI to play a significant role in the future of architectural practice. Yet assessments of immediate benefit were more restrained: 53% described acceleration of the iterative design process as marginal, while 35% described it as high.
The survey suggests a profession that expects AI to matter, while still looking for reliable evidence of where it improves day-to-day architectural work.
Nearly everyone wants rules.
74% agreed that ethical guidelines for AI in architectural design are necessary; 7% disagreed. The leading concerns were intellectual property (40%), quality assurance so that AI-informed decisions do not undermine health and safety (36%) and transparency about when AI has been used (22%).
Why I keep returning to this report
The gap between the 74% using AI for renderings and the roughly one-fifth using it for code, feasibility or layout is the central question. Image generation is where mature off-the-shelf tools are readily available. It is not necessarily where the profession's most costly, repetitive or accountable work sits.
The demand for precision and control, together with the drop in satisfaction once work moves beyond early design, points in the same direction: general tools work best where outputs remain open to interpretation. They struggle where an architect must explain, coordinate and defend a decision.
That boundary is where more specific architectural tools become worth building.