I’ve spent the better part of a decade working on personalization, search, and recommendations at scale, most recently at Netflix, and now in a new chapter at Zoctok. The through line has stayed the same for me: recommendations, search, personalization.But the way I think about building these systems has shifted dramatically over the last couple of years, and that shift is what I want to walk you through.We’ve all heard plenty about large models. Hopefully you’ve also had a large coffee at some point today, because what I want to talk about is a little different. I want to talk about what happens when smaller models start making the bigger decisions.A state gutted its AI law. Then Congress stepped in.America’s AI regulatory landscape just had a month that made legal counsel everywhere reach for stronger coffee. Colorado’s landmark AI Act, once celebrated as the country’s first comprehensive state AI law, was gutted and replaced before it ever took effect.A decade of building prediction systemsLet me set some context. Over the past ten years, I’ve worked on large-scale systems that make billions of decisions every day: prediction infrastructure, search systems, candidate generation, ranking.Early in my career, most of the work centered on improving the predictions themselves. Better ranking models. Better relevance signals. Better signal quality for search.And generally, when predictions improved, the systems improved. That was the loop we were optimizing for.Lately, though, something has changed in how we work. We’re still building increasingly sophisticated models, and we’re still investing heavily in the infrastructure that supports them.But the hardest problems are no longer purely about model quality. They’ve shifted to how we make decisions across technical and organizational layers, and how we connect those dots.With the rise of generative AI and more specialized, smaller models, that shift has only accelerated. We’re moving from building prediction models to building decision systems. And that’s worth talking about.
For expert advice like this straight to your inbox every other Friday, sign up for Pro+ membership.You’ll also get access to 300+ hours of exclusive video content, a complimentary Summit ticket, and so much more.So, what are you waiting for?
Get Pro+
GPS vs autopilotI’m a visual thinker, so here’s how I frame it: GPS versus autopilot.GPS gives you predictions at a first-principles level. Take this route. Avoid this traffic. You’ll arrive at this time. After that, the system stops. You’re still driving. You’re still adapting to traffic, reacting to changes, and making the decisions.Autopilot works differently. It continuously adjusts. It’s taking actions and reasoning on your behalf. It observes conditions, reacts to turbulence, and makes decisions in real time.