David Luan: evaluation

Transcript context

…Interesting. One of the, our investment teams has this idea of services as software. I'm actually giving a talk at NVIDIA GTC about this, but basically software as a service, you're wrapping user productivity in software with agents and services as software is replacing things that, you know, you would ask somebody to do and the software just does it for you. When you think about these use cases, do the users still go in and look at the agent kind of like doing the things and can intervene or like are they totally removed from them? Like the truck thing is like, does the truck just show up or are there people in the middle checking in? I think there's two current flaws in the framing for services as software, or I think what you just said. I think that one of them is like in our experience, as we've been rolling out Adept, the people who actually do the jobs are the most excited about it because they don't go from, I do this job to, I don't do this job. They go from, I do this job for everything, including the shitty rote stuff to I'm a supervisor. And I literally like, it's pretty magical when you watch the thing being used because now it parallelizes a bunch of the things that you had to do sequentially by hand as a human. And you can just click into any one of them and be like, Hey, I want to watch the trajectory that the agent went through to go solve this. And the nice thing about agent execution as opposed to like LLM generations is that a good chunk of the time when the agent fails to execute, it doesn't give you the wrong result. It just fails to execute. And the whole trajectory is just broken and dead and the agent knows it, right? So then those are the ones that the human then goes and solves. And so then they become a troubleshooter. They work on the more challenging stuff. They get way, way more stuff done and they're really excited about it. I think the second piece of it that we've found is our strategy as a company is to always be an augmentation company. And I think one out of principle, that's something we really care about. But two, actually, if you're framing yourself as an augmentation company, you're always going to live in a world where you're solving tasks that are a little too hard for what the model can do today and still needs a human to provide oversight, provide clarifications, provide human feedback. And that's how you build a data flywheel. That's how you actually learn from the smartest humans how to solve things models can't do today. And so I actually think that being an augmentation company forces you to go develop your core AI capabilities faster than someone who's saying, ah, okay, my job is to deliver you a lights off solution for X. Yeah. It's interesting because we've seen two parts of the market. One is we have one company that does agents for SOC analysts. People just don't have them, you know, and just they cannot attract the talent to do it. And similarly, in a software development, you have Copilot, which is the augmentation product, and then you have sweep.dev and you have these products, which they just do the whole thing. I'm really curious to see how that evolves. I agree that today the reliability is so important in the enterprise that they just don't use most of them. Yeah. Yeah. No, that's cool. But it's great to hear the story because I think from the outside, people are like, oh, a dev, they do Act One, they do Persimon, they do Fuyu, they do all this stuff. Yeah, it's just the public stuff.…

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