Built to compound: the marketing engine we run at Tandem

A few weeks ago, one of our AI agents wrote an ad. Another read it, decided it wasn't good enough, and sent it back. By the time it reached me, it had been rewritten three times, and I hadn't touched any of it.

One of the companies I work with as a fractional COO is Tandem, a small, early-stage tech-for-good startup. Our app enables parents and children to co-create personalized, illustrated stories based on their unique interests and preferences. I use Tandem nearly every night with my son, and it has brought so much joy and connection to our relationship over the last few months.

Our challenge is to show the value I see every day to the families who would benefit from it, and offer enough that they want to subscribe. But most of the team is part time, and we need to test a lot of marketing ideas fast, without the budget or headcount a bigger company would have. So we sat down around a whiteboard and sketched what an agentic marketing system that could learn and improve on its own might look like. At the time, it felt like science fiction.

Then I sat down with Claude, and we built it, piece by piece.

It designs a campaign around personas we researched (parents who would pay, never marketing directly to kids), writes the landing pages and the ads with agentic loops that only reach us when they have passed our definition of quality, and launches them. Then it watches how they perform and runs small experiments each day to improve them. It is not running blindly. At every stage, we have built in our own brand materials, context, and knowledge, and we mapped out exactly where human judgment is needed. Nothing goes live until a person signs off. The system does the work. The judgment stays with us.

A big part of what makes it work is our team's own expertise. We spent hours with our marketing lead mapping out what actually makes a campaign work, and built those standards into the system, so it applies them to every ad and landing page, while he spends his time on the strategic work that genuinely needs a human.

It's still early, and it still gets plenty wrong. But it's already let us run several campaigns in the time it used to take to build one, and it's changed how I think about what a small team can do.