Architects are sitting on a data gold mine, and Anthropic and other frontier AI labs can’t get to it

Architecture may be one of AI's most challenging pursuits. Unlike coding or writing, the data necessary to train architecture AI models is not readily available online. Instead, it resides within the servers and physical filing cabinets of individual architecture firms, comprising intricate digital drawings, 3D models, and hand sketches essential to each project.

Presently, none of the major AI labs are pursuing this challenge, leaving it to the firms themselves to harness their data. Architecture firms, both small and large, are actively investing in AI capabilities by recruiting data scientists and machine learning specialists. They are running AI idea competitions, developing bespoke plugins and apps, and even creating custom large language models tailored to their unique style.

This surge of activity has led the industry to recognize that architecture firms possess invaluable data that could help them stand out in an increasingly competitive market. "Everyone has their little gold mine they're sitting on, with all this data of past projects they've done," says Faizan Zaidi, director of design technology at Spectorgroup.

However, the question arises: which firms are willing to build the tools to leverage this data? Four years ago, Matthias Hollwich, cofounder of the 15-person New York-based architecture firm HWKN, was introduced to Dall-E, which marked a significant moment of AI awakening for him and his colleagues. Since then, Hollwich has integrated numerous AI tools into his firm's workflow for office, residential, and hospitality projects.

These tools have revolutionized their approach to design and the final products they create. They now utilize AI for zoning analysis, conceptual designs, renderings, site plans, floor plans, building code compliance, construction documents, and BIM. These tools expedite design concept development, ease complexities in translating initial designs into tangible materials, and enable deeper analysis to better serve users.

HWKN has even built a custom RFP response tool using Claude. Spectorgroup, the architecture firm represented by Zaidi, has also developed custom AI tools to complement their existing design programs. They utilize AI-assisted coding tools to create software that automates tedious tasks, such as tallying conferences rooms, pantries, and desks within floor plans.

While these AI tools have significantly improved efficiency, Zaidi acknowledges that they have not yet been able to fully replace the expertise of human designers, particularly in placing elements like noisy pantries and conference rooms. Overall, architects are rapidly building custom AI tools to stay competitive in the evolving industry, with AI tools becoming an integral part of their design process.