Invest Like the Best with Patrick O'Shaughnessy
· Gokul Rajaram
[00:00:00] S1: Yes. The one thing I think that's going to be truly future-proof is judgment. Why? Because what is the biggest challenge you have when you have a thousand AI engineers writing code? You have the big challenge of AI slop. Every product director I’ve talked to is extremely worried that because you have these engineers running rampant, they're just going to produce lots of code. Which of this code is even valuable? In an era when you can do everything, the question is which of these things matter and you should truly do. On the product side is judgment around what needs to be built and evaluating the output. On the engineer side is evaluating the code, because if you don't understand what the code says, I think you can have AI engineers writing beautiful code that could be wrong, that could have bugs in it, that could be vulnerable. Someone needs to review it and make sure – you have to have human review at some point, especially the critical code that is in the core of your system. And similarly in design, you have to have judgment around, does it make sense in the broader design system? So I think this judgment is the number one thing that humans are going to bring. Era of infinite productivity, the question is what are the things we're productive on, and are we building the right things?
· Patrick O'Shaughnessy
[00:00:00] S0: As you evaluate companies today, build things yourself and just think about the trajectory of these tools. Maybe walk through how someone should think about building an AI application. There's so many people excited about… it feels like a gold rush with this new technology. So many things that we can do that we couldn't do before or things that specific people couldn't do because they weren't technical that they can now do. How should people think about building an application using AI starting today?
· Gokul Rajaram
[00:00:00] S1: First and foremost, you've got to start with a deep and compelling problem. The good news is, there's a tremendous number of deep and compelling problems today, in every vertical and every industry. Why? Because till today, till recently, software was used more as a tool by humans. We finally have software that is agentic in nature, which means it can do the job of people. The question you have to ask is, what industry are there roles of people that are highly paid, that are doing somewhat of a repetitive job, and that can be done by software? Every three months, the answer gets deeper and deeper. You couldn't have told me that a designer's job could be automated by AI six months or nine months ago. You couldn't have told me that an architect's job could be automated by AI. A lawyer's job could be automated by AI. It turns out, increasingly in every vertical, these capabilities are getting better and better. So you want to start with, what industry do you want to be in, and what kind of job do you want to do? Second, you want to target a high-value workflow. You want to target a workflow that is deep, that is complex, and that requires custom data. I think one of the challenges with this whole space is that the models are becoming so good that, if you try to build a company that is light, that is not a hard problem, the foundation model companies are going to eat you. I met with the CIO of a Fortune 500 company a few weeks ago. I was asking him about a few startups I had invested in and worked with. He said, "Look, I don't know why I would use any of these startups. Gemini has an agent builder product, and I also use ChatGPT Enterprise, and they also have an agent builder product. And I have a thousand IT engineers who work for me. They all want to be retrained as AI engineers. So I'm just going to put them using these horizontal tools to build my AI agents. Why do you need any startups?" And so that's the kind of thing you're going to face, that if the CIO of a company, of your target customer, can build what you're building using these agent building tools, you're not going to be successful. So you've got to really go one step ahead of what can be built. Multiple steps ahead.
· Gokul Rajaram
And you've got to extrapolate to where can the capabilities of these agent building products go, and you've got to do something very, very different. So what that means is you've got to have durability, because ultimately as venture capitalists, or even as an entrepreneur, your time horizon can't be building something that lasts for one year. And that's the biggest challenge. It's not building an application, it's building an application that's durable, that basically will last a test of time. And I think there are a few things around durability. One, you need to have ownership of a scarce asset. A scarce asset could be a license of some kind, it could be a regulation of some kind where you have unique insight into it. Second, you might basically own a control point. A control point is a thing that controls how people interact with money or with data. Third, you want to maybe have hardware, which is hard to replace. Fourth, maybe you want to be part of an essential workflow. Fifth, you want to have network effects. You want to think about those things and figure out how after you take on that workflow, you can make it more durable. And finally, I think your ambition has to be to replace the entire system. In other words – increasingly what is going to happen, and I'm seeing this more and more, is every vertical has either a legacy, or somewhat new, what is called a system of record. Which is a system where most of the data is stored for that system. For example, in legal, there's a company called Filevine or another company called Clio. In sales, it's Salesforce. In healthcare, it's Epic. Now, for many years, these companies all had APIs that if you enter that industry, you could build an agent company on top of these APIs. In 2024, things changed. These companies started seeing that these agent companies, AI companies that are being built, they are starting to take on the functionality out of these companies and are treating them like a dumb database. So you started seeing last year that these companies are cutting off access to APIs. Slack has done it most publicly. Slack is owned by Salesforce. They cut off access to Glean, where Glean can no longer access Slack data. And the reason is they don't want Glean to build on top of them and then slowly suck out the value that Slack has.
· Gokul Rajaram
And I'm hearing from other verticals that they're doing one of three things. They're blocking access to APIs. They're offering their own agents for free, bundled. I think that is a great and effective strategy. Or they're charging these AI agent companies to access the data. Just to access data, the API was free. They're saying now it's $2 an API call. They're trying to make the model of these agent companies unviable. I think it's going to be very hard for an end customer to use multiple companies, where you have a system of record and then you have this agent that sometimes doesn't work with it properly. So the agent companies have no option but to also start building and offering a system of record. So every company I know is now trying to figure out how do I build the entire platform, and not just a system that does some workflows. I think last year, everyone was like, "Oh, we can do workflows. We can build what is called a system of action and live on top of the system of record." I don't think that's an option anymore.
· Patrick O'Shaughnessy
[00:00:00] S0: The Slack example is a good one of a last generation software company, which was very big and very successful. One of the most interesting investor questions, and I'm curious for your answer from the perspective of a builder and a technologist, is that the degree to which these horizontal model companies are going to destroy or be very bad for old software companies because over time it will be trivial to spin up your own Slack that has features that you want for your company and it's very reliable in all those same ways that Slack is, and therefore Slack's in a lot of trouble. How do you think about that question? Obviously, public markets seem to think software's in a lot of trouble. The multiples are really, really low. How much would you be worried if you ran a good, solid, but older software company today?