Canadian cultural institutions are racing to embed AI into their operations, and nobody's quite sure why. At the National Summit on Artificial Intelligence and Culture, held in March at the Banff Centre, Walrus editor-in-chief Carmine Starnino found that organizations like the CBC, National Film Board, and Royal Ontario Museum had already woven AI into daily tasks like event planning and financial reporting. The pressure is real. Provincial reps told Starnino that officials are pushing hard to experiment even when the rationale is thin.

One application stands out as genuinely useful. Sim Canada, a production services company, built software that maps film shoot logistics, from truck parking to street closures. An agentic workflow then fills out municipal permit applications automatically. What used to take days now takes minutes. That's the kind of concrete productivity gain that makes AI adoption feel earned, not forced.

But the summit's dominant mood wasn't excitement. Starnino describes an "uncomfortable third state" where attendees adopt AI out of collective fear, not conviction. Arts administrators spoke about the stress of using AI for core duties like grant writing, even when it produces hallucinated details and confident errors. One woman said her boss bragged that AI would "10X" her output. The questions piled up. What about job losses? Who protects the intellectual output? Just as an AI-run store recently struggled with hiring and inventory, cultural workers worry about losing their place in the production line.

The federal government doesn't seem interested in these concerns. AI czar Evan Solomon took the stage to talk about scale and compute, framing AI as an industrial race for capacity. The misalignment was jarring. Cultural institutions are worried about surviving an AI-saturated environment, and the people in charge are talking about building more data centres. Nobody seemed to have a plan for what comes next. Just a shared sense that culture will be asked to catch up later, after the race is already run.