Alexander Embiricos: belief

Transcript context

…Just a couple more questions. One, we touched on this a little bit, as AI does more and more coding, there's always this question of, "Should I learn to code and why should I spend time doing this sort of thing?" For people that are trying to figure out what to do with their career, especially if they're into software engineering computer science, do you think there's specific elements of computer science that are more and more important to lean into, maybe things they don't need to worry about? What do you think people should be leaning into skill-wise as this becomes more and more of a thing in our workplace? I think there's a couple angles you could go at this from. Well, the easiest one to think of at least is just be a doer of things. I think that with coding agents getting better and better over time, it's just what you can do as even someone in college or a new grad is just so much more than what that was before. And so I think you just want to be taking advantage of that. And definitely when I'm looking at hiring folks who are earlier career, it's definitely something that I think about is how productive are they using the latest tools? They should be super productive. And if you think of it in that way, they actually have less of a handicap than before versus a more senior career person because the divide is actually getting smaller because they've got these amazing coding agents now. So that's one thing, which is, I guess the advice is just learn about whatever you want, but just make sure you spend time doing things, not just fulfilling homework assignments, I guess. I think the other side of it though is that it's still deeply worth understanding what makes a good overall software system. So I still think that skills, like really strong systems engineering skills, or even really effective communication and collaboration with your team, skills like that I think are important or are going to continue to matter for quite some time. I don't think it's going to be all of a sudden the AI coding agents are just able to build perfect systems without your help. I think it's going to look much more gradual where it's like, okay, we have these AI coding agents, they're able to validate their work. It's still important. For example, I'm thinking of an engineer who was working on Atlas, since we were talking about it, he set up Codex so that it can verify its own work, which is a little bit non-trivial because of the nature of the Atlas project. So the way that he did that was he actually prompted Codex like, "Hey, why can't you verify your work? Fix it," and did that on a loop. And so you still, at various phases, are going to want a human in the loop to help configure the coding agent to be effective. So I think you still want to be able to reason about that. So maybe it's less important that you can type really fast and you understand exactly how to write... Not that anyone writes a 4H loop or something, or you don't need to know how to implement a specific algorithm. But I think you need to be able to reason about the different systems and what makes a software engineering team effective. So I think that's the other really important thing. nt a specific algorithm. But I think you need to be able to reason about the different systems and what makes a software engineering team effective. So I think that's the other really important thing. Then maybe the last angle that you could take is, I think if you're on the frontier of knowledge for a given thing, I still think that's deeply interesting to go down, partially because that knowledge is still going to be... Agents aren't going to be as good at that, but also partially because I think that by trying to advance the frontier of a specific thing, you'll actually end up being forced to take advantage of coding agents and using them to accelerate your own workflow as you go.…

Stored transcript either side of the excerpt. The highlighted words are the published quote; the surrounding text is unedited source, never generated.