00:14:53 · Srinivas Narayanan, Aditya Agarwal
So you have to make really informed and smart decisions about, okay, what systems are okay buying and not building. Right? So it's just a combination of all of these things that we had to go through. [00:15:04] S1: I think both of them are interesting points. Right? I actually want to talk about the first one first. You know, at early stages, often, there's no personal belief that the team is the company. Right? And there's been a lot written about maybe the drama of people leaving some of the labs. But I think what's more interesting is why do people come to these labs, right? What is about the culture and the mission and the ability to make an impact? And how do you communicate and get the right kind of people? I'm curious about that. Because when I look at the kind of people who are there, it's completely mind blowing across the board. And I'm not only talking about like, OpenAI, Anthropic, but even if I look at the world model companies across the board, I think the quality of people who are who are going to all of these places are amazing. So what what has been the fundamental unlock? Right? And specifically in OpenAI's context, what really worked?
00:15:58 · Srinivas Narayanan
[00:15:58] S0: Yeah. I think the most important thing is building an extremely highly talent and steam and small teams that are... Where everybody respects each other for what they bring to the table. And so you have to just put a lot of effort into your hiring process, hiring culture, who do you bring in. And if something doesn't work out, you know, it's not that that person isn't great, but it's just not a great fit for the company. Be willing to make those decisions, hard decisions if things aren't working out, and just maniacally focusing on what's working. I think you also have to have a good understanding of what is right for your company. Right? For OpenAI, there was a whole bunch of things about, you know, really deep passion for why you're here, the passion for AI, the ability to move really quickly, the ability to make decisions in uncertainty, the ability to have a lot of agency, you know, having curiosity. So there's a bunch of things that, you know, we really, really... Ability to collaborate really well with others. And, you know, the profile that I would say really worked for OpenAI is very, very entrepreneurial people. Sometimes you hire for what is in a resume. That is not necessarily what is right for the company. You know, I think just being able to, one, incredibly smart, but also being extremely entrepreneurial and getting things done very quickly, these are extremely important. I think that is sort of true for a lot of startups. So I think we were incredibly lucky. We're also... I mean, you just look at the impact AI is having, especially OpenAI and other labs are having. I think it also brought a lot of people into the table. So we were very lucky in that way, which is that everybody wanted to work there. So we didn't necessarily have a problem of trying to attract. We had very good people who wanted to come here. And then, of course, you know, you have friends, you have referrals, etcetera. But you have to spend a lot of time on figuring out what's the right set of values, what's the right sort of people you want for your company. And every company is unique. And so that's very, very something you have to craft a lot.
00:17:55 · Aditya Agarwal
[00:17:55] S1: Got that. The other part that you talked about, Fortune 500 CEOs and companies adopting AI. Right? And a bunch of our founders and both founders here and globally are building for the largest scale adoption whenever it happens, right? What is your view on the rate at which adoption has been happening in Fortune 500? And especially as a bunch of us have moved our workflows to agents and some of us are taking to it like fish at sea. Right? And I'm just curious, what are the specific engineering problems that need to be solved for some of these Fortune 500 CEOs and CFOs to believe that, hey. You know what? It's time to hand over some of the jobs to be done here or maybe go completely autonomous in some areas.
00:18:40 · Srinivas Narayanan
[00:18:40] S0: I mean, as we all know, we are seeing tremendous adoption in areas like coding where the the output is fairly verifiable. I think in more complex or maybe in other functions where things are less verifiable, it's harder. It's harder for a whole bunch of reasons, especially in large companies. One, how do you get the right context to the model? Because a lot of the way people make decisions at companies, there's institutional knowledge. Things aren't written down clearly. Yeah. So you can't really just get an AI to go and do that. Right? Like, an analogy that I've been saying is it's it's a little bit like hiring an extremely smart person into your company, them having almost a blindfold. You know, you can't talk to anybody else. You don't know what is where, but you have to go do your job. You're incredibly smart, but most people will fail. Right? You won't know how to do. The way you onboard a person in a company is by having them talk to others, understand how a company works. Yes. It's all building that institutional context. So that, I think, is not trivial to do today. This is why you see enterprise adoption being slower. I think that's one big challenge we have to do. Being able to learn on the job. You know? If you make one mistake as a new person in a company, people say, oh, don't do that. Do it the other way. And then we register that. Right? So how do you turn that into a repeatable and trustworthy process for an AI agent or job? That's the second problem we have to solve, just learning on the job. I think the third problem we have to solve is is security. Mhmm. I think there is a fair amount of concern about how agents can go rogue. Can they do all these things that, you know, do I trust them? So we really, really have to solve the trust problem is to pull guardrails around them, to actually have deterministic controls on what your agent can do and cannot do. And maybe there are certain risky decisions that you're not comfortable offloading to an agent, but you can start small. And I think gradually, the risk profile for what you let an agent do will probably grow... Will grow over time.
00:20:31 · Aditya Agarwal, Srinivas Narayanan
[00:20:31] S1: Yeah. It almost also feels like, you know, you need to rewire the entire org structure because both incentives, motivations, and the way some of these orgs have worked historically, even holding information, holding holding power. And all of those things play out in various ways where a complete agent take Yeah. And a human workplace Yeah. I still feel feel some distance away. Right? [00:21:00] S0: That's true. I know I was saying this earlier. Right? I think there's a famous quote about, you know, we humans have paleolithic emotions, medieval institutions, and godlike technology. And it's sort of true. All the organizations are designed for humans doing things during the last hundred years. I think if you want to... So this is why you're seeing individual productivity is changing because of use of assisted tools. But the way most organizations work is actually how work gets drawn across teams, And that has a whole bunch of business processes and customs and things that you're used to and designed for humans being at the center. You have to redesign them for AI. I don't think that transformation will happen before that. I think there's good analogies in how electricity panned out as well. Factories had to be redesigned to actually accommodate electricity. Right? Like, couldn't just... You know, they weren't seeing the gains when electricity first happened until they completely rewired the the... How the factories were set up. So I think there is some process for that also, which all, I think, companies have to go through.
00:21:59 · Aditya Agarwal
[00:21:59] S1: One of the rewiring organizational rewiring that a lot of people are worried about is the fact that most orgs historically have worked with you referred to it earlier. New people come in. They go into this apprenticeship model, right? And they learn and they become effective. And just seems that today with this chunk of intelligence available, that apprenticeship model seems to be at risk, right? And a lot of younger folks are worried about what a meaningful career means for them, right? And this is more a personal curiosity as a parent. I'm just wondering that, and even for someone who talks to a lot of people who are early in their careers, how should people think about this specific access?