I kept gravitating toward the systems underneath — then I turned to AI.

I studied Communications and got an MS in Public Relations, thinking I’d build in comms or marketing. And I did, for a while.

I spent three years as a media planner in New York, buying across major ad platforms for e-commerce and DTC brands. I got good at it, but I got bored of optimizing inside someone else’s system.

So I joined TikTok in 2021 and spent the next four years moving from product operations to product marketing to product management. I kept getting closer to the engine underneath the banners, not just the slogans on the surface.

That’s the throughline, if you’re looking for one: I keep gravitating toward the systems underneath things. How the machine works, not just what it produces. And I keep ending up in the seat between technical teams and everyone else, translating what the system does into something other people can act on.

Now at a marketing measurement startup, I own product operations for incrementality-driven attribution and MMM, and I build the systems that keep the org moving — knowledge architecture, CS enablement, pipeline automation, release comms, and more. More and more, those systems start in AI.

I got into AI out of frustration: the tools I had couldn’t do what I needed. When the official integration fell short, I built a 50-tool MCP server and started automating the daily digests, ticket workflows, and account health checks that were slowing me down.

That’s what pushed me toward a more AI-native way of working — not as a buzzword, but as a way to make work start, move, and finish inside AI. I wasn’t trying to build AI systems for the sake of it. I was trying to solve my own broken workflows faster than the roadmap could, and make execution actually better — not just for me, but for the team that had to live with it.

The bottleneck in AI-powered work isn’t the tool. It’s whether the person directing it has a calibrated sense of what good looks like.

AI is a parallel execution layer I direct at the architecture level, not the sentence level. My job is taste, judgment, source weighting, and knowing when to override. The more precisely I can specify what good looks like upfront, the less I need to review at the end.

I think the organizations that figure this out early — not just the tools, but the new operating model — will have a durable advantage. And I think most won’t figure it out through top-down adoption programs. It’ll happen because a few high-agency individuals explored first and built the patterns everyone else eventually copied.

There’s a classic idea in communication research — Joseph Walther’s Social Information Processing (SIP) theory — that says relationships built through computer-mediated communication can ultimately reach the same level of closeness and trust as face-to-face interaction; the difference is speed, because you have fewer social cues per moment and need more exchanges to get to the same depth.

In the AI era, I think that “extra time” flips from a limitation into a design choice: when the work is mediated (and some of it delegated) through systems — AI copilots, automated workflows, and well-structured knowledge — people who can direct those systems well can protect attention, raise quality, and reduce the noise that usually consumes our days.

The payoff isn’t just productivity. It’s reclaimed bandwidth for the kind of human connection that can’t be automated: clearer collaboration, better judgment calls, and relationships built with presence instead of interruption.