Architects in Canada are encountering a new participant in client meetings: generative AI. A feature in the October 2026 issue of Canadian Architect describes clients editing early renderings with AI tools and returning with their own altered versions, and finds the profession divided on whether that helps.
Uros Novakovic, co-founder of Office OU, says his team often shares early Revit renderings with clients on Miro during video calls. Recently, clients have arrived having already reworked those images with AI. He compares it to clients bringing photos of existing buildings as precedents, except that AI lets them apply the precedent directly to the design, and he finds the practice largely constructive.
When an AI edit ignores the building code
Heather Dubbeldam, principal and creative director of Dubbeldam Architecture + Design, is less enthusiastic. She recalls a client who asked an AI to change a single element of a rendering, producing an image that made little architectural sense. Her team then had to spend extra time explaining why the change did not work aesthetically and, more importantly, why it would not meet code.
She also notes that AI image tools tend to regenerate a scene rather than refine it. Asking for better trees or a nicer street view can result in a different facade material. Dubbeldam worries that clients will come to expect every part of design to move at the speed of a prompt, when good work still takes weeks rather than days.
Where AI is already useful in practice
Most of the architects interviewed use AI mainly for routine office work. Evan Saskin of Blue Lion Building describes using ChatGPT to work with spreadsheets, documents, accounting software and email, and runs zoning calculations for Toronto infill sites by moving between the city’s bylaw map, a spreadsheet and a chat window. Janna Levitt of LGA Architectural Partners uses AI to check RFP documents against contracts, saying it gets about 50 to 60 percent of the way at first and 75 to 80 percent with better prompts, though everything still needs checking.
Code compliance may be the most disruptive area. Novakovic asks Anthropic’s Claude code questions, while acknowledging it does not always answer correctly. Toronto firm Trax has offered a building code co-pilot since 2023 that answers questions on provincial and national codes with citations and drafts deficiency letters, and in some US jurisdictions tools such as Symbium and CivCheck already help issue permits for simple projects.
Builders remain cautious. Oliver David Krieg of mass timber firm Intelligent City says construction demands a precision that large language models cannot yet deliver without layers of human control. Contractor software founder Luigi La Corte expects AI to close scope gaps in drawings, but says value engineering will be hard to automate. For now, the industry still lacks an integrated platform that carries a design from drawings to approvals, bids and orders.
Leave a comment