Brands Now Sell to Two Audiences: Humans and AI Agents

By 2026, an AI agent can photograph your empty living room, match furniture to your budget and style, and complete the purchase before a human ever opens a retailer's product page. Toby Coulthard, chief product and growth officer at Jacquard, an AI messaging platform, says that shift splits every brand's audience in two — and most marketing teams have optimized for only one of them.

"Instead of just selling to one customer base — humans — brands are now selling to two," Coulthard wrote in a Crunchbase News essay. The second audience, AI agents, decides on different criteria than people do. Emotive copy marketers spent years crafting may be broken down and reconstituted by a machine before a shopper sees it.

The claim is directional, not measured. Coulthard cites no adoption data for agent-mediated purchasing, and he concedes the point himself: "No one knows how fast AI purchasing will scale." Humans, he notes, still do most of the reading and buying today. What follows are his recommendations for teams preparing for both audiences.

Who gains an advantage?

New startups, in Coulthard's framing. Established brands must retrofit decades of copy, brand guidelines and approval processes to become agent-ready. Founders building now can write copy legible to humans and machines from day one. The cost asymmetry is the argument: building these decisions in early is cheap; retrofitting later is expensive.

What should teams actually do?

Coulthard's tactics apply across stages, but several assume early-stage companies:

• Write the brand voice down. In most early-stage companies the voice exists only in the founder's head. That works when a handful of human writers know the founder. It fails when the writer — or the reader — is a machine. AI tools follow explicit rules and drift without them.
• Document the negatives. What a brand will not say matters as much as what it will. List banned words and prohibited claims before humans or machines produce copy at scale.
• Treat product descriptions as brand assets, not admin. In an agent-mediated world, descriptions may be the most-read copy a brand owns. Two agents recommending the same chair to two buyers may each tweak the description; the source copy must carry the brand intact through that transformation.
• Test the second reader in email. Gmail and Apple now generate AI summaries that appear above the email body and may be read instead of the message itself. Before sending, Coulthard advises, check what the summary says the email is about. Does the offer survive? Does the differentiator survive? The check needs no tooling and takes little time.
• Measure outcomes, not volume. Teams that gauge success by content produced will find that metric meaningless as AI makes production cheap. Conversions and revenue are the metrics that still show whether marketing works.

What does this change for marketers?

The workflow change is concrete and available today: the email-summary test requires zero budget, and the voice documentation is a writing exercise. The larger bet — agent-mediated purchasing at scale — remains unquantified. Coulthard, who has spent years at Jacquard building AI agents that write brand content, says agents appearing on the reading side of the process is what makes explicit rules newly urgent.

His caution cuts both ways. Marketers should not abandon the human in the rush toward agents, he argues, because every agent ultimately serves a person's preferences, and personalization will only get more granular. The recommendation is simultaneous optimization for both audiences, not replacement of one with the other.

What happens next?

Coulthard expects retail to look very different in five to 10 years, without committing to a mechanism or timeline. His closing frame is an investment thesis: brands that encode dual-audience practices now pay a small cost, and brands that wait pay for a retrofit. How fast the second audience grows — and whether it decides before the first one looks — is the variable every marketing team will be measuring against.