Even as AI agents write more code, engineers must retain deep understanding of how systems work so they can judge quality and fix things when they break, not just the ability to write code in a given language.
Stone argues that while AI may reduce the need to hand-write code, engineers must retain deep understanding of how systems work so they can judge quality and fix things when they break. ✦ AI generated
Elizabeth Stone · Lenny's Podcast · 2026-07-19 · original ↗
starts at this moment · 56:28
“Where do you think engineering goes in the I don't know, 5, 10 years? Do you think people need to still understand code? Or do you think there's this abstraction layer that sits on top where you don't even have to learn C++, Java, Python, whatever?”
I think there's a difference between being able to write lines of code in a particular language like Python or C++ and understanding how code, computer systems, products work. And I don't think the latter is going away. Because if we trusted agents to know all the languages and write all the code, we're not going to know why is something Is it a good product? Is it a bad product? Is it working as we expected when it doesn't?
verbatim transcript · starts at 56:28
56:28the I don't know, 5, 10 years? Do you think people need to still understand code? Or do you think there's this abstraction layer that sits on top where you don't even have to learn C++, Java, Python, whatever? >> I think there's a difference between being able to write lines of code in a particular language like Python or C++ and understanding how code, computer systems, products work. And I don't think the latter is going
56:59away. Because if we trusted agents to know all the languages and write all the code, we're not going to know why is something Is it a good product? Is it a bad product? Is it working as we expected when it doesn't? Like I mentioned earlier, we take a lot of risk. We fail fast, we recover fast. That requires an understanding of how are these systems working. I might use
57:21an agent to help me understand those things, help me detect an anomaly or something that's broken faster and triage it, but I still need to have a fluency of like, what is this thing that we're building and how does it work? So I know if it's good and I know how to fix it. I don't know I I hope that doesn't go away cuz it you know, that that's like a
57:41how do we make the world a better place through the stuff that we're building? I think requires some understanding of what we've built. >> What I'm hearing which it makes sense is you may not have to write the code but you have to understand it and what's happening. But it's so much harder to just as a person not writing it to actually you know, have that instilled
57:58in you. >> I think that's one of the the things that the learning curve is very steep on right now. So looking at some of the code that some of these models or agents are writing they're very hard to follow. It's like I know I'm getting better performance from this but I have no idea why and if this thing breaks I'm going to have no idea how to fix it.
58:17That that's makes me uncomfortable. You know, maybe that's because I'm still on that learning curve of like how do we operate in that world? Like what's the set of tests or rationalization and understanding that we need to have to get comfortable with it? But at first glance it looks very unfamiliar and very unsettling. So I think engineering over time will evolve to be comfortable with that and have fluency in it and know how
- ·Knowing a language like Python or C++ differs from understanding systems
- ·That deeper understanding isn't going away, says Stone
- ·Agents may write code, but can't judge if it's good
- ·Engineers must judge if a product actually works well
- ·Need to diagnose and fix things when they break
- ·Trusting agents blindly risks losing this judgment