Toby Coulthard: write brand voice for AI agents, not just humans

Coulthard's starting point is a shift in how purchases begin. Ten years ago a new homeowner might search Google or consult interior design magazines; in 2026, he writes, they could photograph an empty room, upload it to an AI agent with a budget and style, and even let the agent buy. That makes the second audience — AI agents — a real buyer, and one that judges on different criteria than people.

His caution is that this does not make the human irrelevant. Each agent still serves a person with their own preferences, so he frames the task as optimizing for both audiences at once. Here he sees an asymmetry: established brands must retrofit decades of copy, guidelines and approval processes, while newer founders can bake legibility to humans and agents in from the start — which is why he calls retrofitting later expensive.

The practical steps follow from that. A brand voice that exists only in the founder's head works for a handful of human writers but not for a machine, which drifts without explicit rules — including rules about what the brand should never sound like. Product descriptions, meanwhile, move from administrative text to possibly the most-read brand copy a company owns, since agents read them when recommending or purchasing; if two personalized agents surface the same chair, the source description must carry the brand intact. Coulthard also ties measurement to this: as AI makes content production easier, counting output becomes meaningless, and conversions and revenue are the metrics that show whether marketing is doing its job.