Between 2014 and 2024, the gap between a new AI technique appearing in the literature and its first use in architecture shrank from 62 years to 2.5. This finding came from a PRISMA review of 161 papers by Roh & Lee (2025) in Automation in Construction. They observed how nearly 70% of that activity sits in the schematic design phase, a field where AI has moved into the early stage very quickly.
Whether the theory can be turned into actual practice is another matter. Yiannoudes (2025) scored 42 deep-learning studies against a workflow maturity rubric and found that very few produce outputs that integrate with CAD or BIM. Most are raster images, pipelines are fragmented, and most prototypes need custom code to run. The productivity gains exist, but inside a narrow lane.
The effect on designers themselves is uneven. Hou et al. (2025) in Information Systems Research ran lab and field experiments and found that AI helps everyone brainstorm, but when it comes to execution, novices still benefit while experienced designers slow down without getting better results. The tool clashes with the routines experts have already built.
At the level of the field, the cost shows up elsewhere. Doshi & Hauser (2024) in Science Advances found that writers using AI produced stories judged more creative, but also more similar to each other. Individually better, collectively narrower. The same mechanism is plausible at work in architecture. Models trained on similar datasets tend to converge on similar visual vocabularies.
Where the technical work is genuinely advancing, the authors say so carefully. Zeng et al. (2024) in Automation in Construction present a diffusion model that passes a Turing test against human experts on floor plans. Still questions around controllability remain. Human contribution and tailoring improve outputs overall. A paradox however emerges. Diffusion model power resides in its stochastic nature while it becomes useful only to the degree it can be constrained.
The direction the five papers point towards is fairly consistent. AI is fast, measurably creative, and technically impressive at the front of the design process. It is also workflow-immature, unhelpful to experts in later phases, and quietly homogenising at the level of the profession.
Want to know more about the effects AI is having on the world of design and architecture? Look at the startups we featured this week on AI World!