“A better model doesn't make you have a better product. You can have a really powerful model and a horrible experience. And we have seen that a smaller model trained on very clean data that you own can give much better and much safer results.”
Product Impact Podcast | Secrets to unlocking the value of AI · Product & Design · October 2026
Context
Asked whether recent AI gains come more from stronger models or from better harnesses and context, Paz says it's some of both. For designers, though, she argues the leverage is in the context and harness around the model.
Transcript
…into really good instructions and rules for the agents to work with. So kind of learning a little bit of the technical side can help you to do that. Saying that I always work with engineering because it can get technical quite easily and I'm not that fluent, but we're all in a learning journey.
…to write explicit rules on what to do. I had to clear it all the time. All that's gone. Now, there's been a lot of gains, obviously. Do you attribute it more to the models getting more powerful? Or is it that everybody's building better harnesses and better context that goes into those harnesses?
I mean, it's a little bit of both. Models had gotten better. Like, you know, the reasoning models we have right now are not the ones we had a few months back and they are much more capable. But having more, like a better model doesn't make you have a better product. You can still have a really powerful model and have a horrible experience. And we have seen, in fact, that if you have a smaller model that you had trained with very clean data that you own, it can, I mean, the results can be so much better and so much safer for sure. So the model doesn't, you want to go with a good One that is like the good fit for the problem you're trying to solve. But I think the harness and the context around it, at least for designers, is still so juicy. And there's so much we can do that because it's truly every piece you put there is going to shape the output and the behavior that the model is going to give you.
So let's talk then about context. I think people understand context a little bit better, but let's put it into another term or like maybe an example that will be more meaningful for people. So let's say, for example, you live in Spain. Right? I think we're all a little jealous of that. So help…
Oh, that's fun. I mean, I have an unfair advantage because I'm a Spaniard and I have lived in the US for many years, so I can like more or less navigate both contexts. But something that I will say is like, I live in Sevilla now, which is this gorgeous city in the south of Spain. And we have many…
And now I'm just imagining Europeans coming to the US and ordering eggnog. I
More from Product Impact Podcast | Secrets to unlocking the value of AI
