IZ Ivan Zhao Ivan Zhao not now, my bread's in the oven

“The deeper level is: does the human have the understanding? Do you understand what's going on? If a team of humans builds a medium-plus-size code base, after a week you sort of don't understand what's going on. I don't think we should surrender our understanding. Definitely not without putting up a fight. If anything, we should try to figure out how to create better abstractions, better tools and better visualization to understand things.”

AI & I · AI Builders · November 2026

“The deeper level is: does the human have the understanding? Do you understand what's going on? If a team of humans builds a medium-plus-size code base, after a week you sort of don't understand what's going on. I don't think we should surrender our understanding. Definitely not without putting up a fight. If anything, we should try to figure out how to create better abstractions, better tools and better visualization to understand things.” — Ivan Zhao, AI & I

Context

Dan raised the 'one more prompt' pull of late-night vibe coding and the idea of having the AI quiz you before it builds. Ivan says the real issue is whether people still understand the systems their agents build.

Transcript

AI & I Around 22:30 into the episode
Ivan Zhao

…after 10 minutes, you got grossed out. I got grossed out. It feels like that a little bit. So it feels like, yes, there's a tremendous amount of productivity gained from agents, from AI, but what other second-order effects can we at least think about a little bit while we're building more to it?

Dan Shipper

…know that we made this thing that makes it really easy to build stuff, and that's awesome. And we know that when it's 11 at night, you can probably just use it to rip tokens that mean nothing and you're just sucked into your phone. How do you think about how to deal with those second-order effects?

Ivan Zhao

Depends on whatever you're building a chat product, maybe just bake this into your system. I think that's too superficial level. The deeper level is like, does the human have the understanding? Like, do you understand what's going on? Like, if you usually build a system or if a team of human medium plus size code base, after a week, you sort of don't understand what's going on. I don't think we should surrender our understanding. Definitely not without putting up a fight. If anything, we should try to figure out how to create better abstractions, better tools, and better visualization to understand things. That's entirely the purpose of science. Math, create better shorthand to understand the world around us. Agent can do a lot for us, but you cannot have understanding. It's difficult, right? I think more software AI toolmakers should think about how do you help people to gain the understanding of what those systems are doing for us.

Dan Shipper

Have you found anything that works for you to make sure that you're locked into understanding how it works?

Ivan Zhao

I can share some tips our teams are using. It's also almost ironic, right? So like think about the way of a coding agent. Clockhold become really popular. Everybody talked into their own little black box to build things. And using GitHub as a way to merge things, then you realize, holy shit, we…

Dan Shipper

So we have time for a few questions. So if you have a question, raise your hand and right in the back over there. Hey, thank you. I wanted to talk about and ask a question about structure versus unstructured. Because the tradition that you're talking about, I think of it as being about comparing or…

Speaker names come from our own diarization. The position is estimated from where the line sits in the episode.

More from AI & I