Paul Lam

For seven years I led engineering at Motiva, building machine learning on top of Oracle Eloqua. Once, about 1,000 contacts vanished from a campaign. The reason was sitting in Eloqua's interface, but its API never told us. The platform knew. We didn't.

We spent years joining systems, finding missing context, and putting it where people could use it. But we had to know what we were looking for and build each connection in advance.

Now I'm curious whether agents can help more people take on problems where a bad answer is worse than none. Agents can look across data and methods while we're still working out what the problem is, surface connections we would have missed, and turn one big judgment call into smaller loops of trying and checking. But we still steer, and we still judge whether what comes back is honest.

I've lived in Japan since 2018, where I bake sourdough without keeping a starter and travel with my family to old country inns—ryokan 旅館.