The Mastermind Behind Palantir's Operations: Shyam Sankar CTO (December 2023)

No one wants to make mistakes. No one wants to be wrong. I wish we could all be right all the time, but I have screwed up so many things at this company over 18 years. If the standard of surviving here was something like you can never be wrong, or you must optically be seen to be making every decision in a way that it will never blow back on you, this company would not exist. I would be screwed. It's not a way to actually make the world better. At the time, it was 12 people, and they wanted to work on problems of national security. My uncle was a victim of the 2006 coordinated train bombings in Bombay. I was very much... and I was in Upstate New York during 9/11. I think that was a seminal moment for many, many folks of my generation. I actually tried to drop out of school to join the CIA, but I think everyone had that idea at that time. I didn't even have a college degree, so they didn't quite respond to my electronic application. I would have much rather failed, and it certainly seemed in 2005, Palantir... it wasn't at all clear how it was going to work, but I would have much rather failed working on something that I thought was so important. We're just committed to building the right product, and I don't want that to sound naive, because actually I think it took us longer and was harder because we didn't know how to sell to government. There were all these things that were kind of zero to one. There was no one else to really learn from. All the people who really knew it were not of the technology world and couldn't get it to work in our context anyways, so we had to flail our way through, feeling it out from first principles and developing a new playbook that was going to work. I think a playbook that subsequently has been really valuable to the rest.

The way the government really wants to buy everything is to pay you by the hour to build it, so some assessment of who should be doing it, but fundamentally it's a labor model. It's a labor model where they really want, through cost accounting and audits, to look at your actual costs, they will build in overhead rates, they will then say this is how much profit you're allowed to have. It's kind of actually pretty communist for a country whose entire economic engine has been powered by capitalism. As a monopsonist buyer, you can say, why is this? It's not totally crazy. Assume you're the only buyer for a product in the world. You're the only buyer for the aircraft carrier. How should I compensate someone who's building an aircraft carrier for me? What's a fair price to pay? The way that they solved this, really starting in the last century, was thinking about cost plus. It seems fair that it costs you something, you have to pay people, you have to buy materials, and I'll give you some profit on top of that. Now, this has obvious consequences in terms of what the long-term incentives end up being. Why would you ever invest in innovation? To the extent the government wants you to innovate, they actually pay you the money to do the R&D. You're not putting your money into it, really. You're putting their money into it. As the world transitioned from a place where all of the R&D was done by the government in the 50s and 60s, obviously the preponderance of R&D is done by the commercial world, this model makes no sense. You really want to leverage the beauty of what we've really pioneered in the Valley: this high-risk, high-return approach. Technology is supposed to make things cheaper. Somehow that doesn't happen in government. To the extent things do get cheaper, they get cheaper because they're benefiting from the massive tailwinds from the commercial markets' R&D that then subsequently makes its way to government. The great opportunity, and really the playbook, was: look, I'm not going to let you pay me by the hour. I'm going to refuse to accept any revenue that's services-based from you. Instead, I'm going to build a product, and I'm going to keep investing in the product. That means I have to go sell a part of my company to finance that. I have to go raise money to do this. But this product is going to be so good you're going to want to buy it, and when you buy it, you're going to buy it on a license fee basis. You're going to pay some sort of subscription software as a service. I know to the Silicon Valley lessers this is like, obviously, what are you talking about? But there was a lot of brain damage to getting the government to this model. Even now, we made a huge amount of headway, but the low entropy, the lowest path of least resistance, is to try to pay someone by the hour.

Land and expand doesn't work the way it does for commercial SaaS companies. What you're trying to do is expand the product surface area, as opposed to the license surface area. I'm trying to get more users. It's not sales-driven expansion; it's actually product-driven expansion. If you think about how do I connect up more of what this customer is doing, I've landed with this capability. How do I expand to lateral capabilities that are valuable to them? Many of those capabilities you're not going to think up in Palo Alto eating strawberries. You need to be in Djibouti, or in Kandahar, or in Balad, or in Ramstein, sitting with the users to understand what's next. Because that truth exists. The truth is out there. It's not in some document somewhere. It's not in DC. The deep understanding of what's missing that you can build true stickiness around, that's one dimension. The other dimension of this is the reason you have to be there is, how are you to your margin optimization? It doesn't actually come from lock-in or just rent collection. It actually comes from understanding how do I continuously productize and create the efficiency in how I deliver what I deliver. There are unique constraints in this environment. You don't have one big multi-tenant or three big multi-tenant geographic instances of your software that you're running on the public internet. You're delivering software into air-gapped environments that are tightly controlled and constrained, and people need clearances to access them. You need to have some sort of product methodology to invest in the software infrastructure. You have your software, then you have the infrastructure below that that is essentially an army of robots that manages your software in production, because otherwise you're going to end up deploying a team of 20 people to go manage your software in these environments. You're going to have a really hard time getting to scale. It also lets us be a little bit more critical of requirements. The requirements are the things that must... we've done this analysis, we said these things must exist, and we're going to build all of them. Well, clearly requirements change. The only thing you know for sure is the requirements are wrong. How do we enable our program managers to take more risk and say, I think these requirements are more important than those requirements, and I think there are requirements they haven't even thought of that really matter? Because at the end of the day, you got to taste the steak. You got to feel the product, use the product, and that's the only way you're going to know if it's good or not. Can you imagine if the Valley didn't have something like that with its buyers? That the way every new product was going to be evaluated was through a historical list of requirements that someone else came up with and a paper PowerPoint exercise to see which product someone was going to buy? No, we innovate because we fight with our technology in the customer's office. Steve Jobs has this great lost interview about how companies get confused. When they're trying to replicate their success, they look at the process they went through to deliver their first success, and then they follow the process, and it doesn't work. Because actually, the first and most important thing is the content. It's the creativity. It's the substance. That substance rarely... and as he points out, people who are good at content tend to be a pain in the ass to manage. They're a pain in the butt, I think was his quote. But you put up with them because the content is just so good. I think there's this way in which process as a medicine is okay. Can you have zero process? No. But if you hold process up as the deity, if process ends up being what is actually an opioid, it's very bad. It's like, we just need to follow the process and we'll get to the outcome. That's almost never correct. When the process becomes more important than the content, you're going to lose. A lot of this comes up like, I wish the interaction could be less frictionful. I wish there was more predictability. I wish there were all sorts of things which are just not true in the real world. Yes, we all wish the world could be different, but when you honor what the world is, you kind of recognize how important it is to have the right rock band. There's no, hey, follow this process and we're going to get there. To put it this way: you can get really far with incredible content and bad process. You're going to get nowhere with incredible process and bad content. I feel Palantir is this company that keeps reinventing itself, which is a whole other discussion we can get to. But in those moments of reinvention, there's just content. There's just the crazy content and your commitment to it and making it better. Then you figure out how to get it to scale. Sometimes maybe getting things to scale is itself not scalable, which I don't mean as a critique. It's more like first you got to figure out the substance, and then you work really hard every day to get it to scale. There's an old Silicon Valley saying: there's the first 80% and the second 80%. Maybe there's a little bit of this where it's like, yeah, the first part of the content is 80% of the work, and then getting it to scale through process is 80% of the work. We've kind of constrained our cultures and our companies because we've leaned into the opioid of structure essentially. Like, I'm going to structure my company as a factory. Here are all the roles and the levels I have, and here's the org chart, and it's all very clean and laid out. If you play by these rules, you can figure out how to navigate. I think one of the challenges is that really talented people are highly asymmetric. You're much better off designing a role around who they are rather than saying, what role can I slot you into? It's almost definitionally none of those roles are going to fit them. It's going to put them in a position of relative weakness. They're going to blow something up. They're going to feel like they have to be good at something they're never going to be good at. It's a total evisceration of their potential. I don't get Dalí to paint better by yelling at him to paint more like Monet. Each one of these individuals is an individual. I'm going to go on a journey to craft a role around who they are, and I'm going to make it okay for them to flex their role to meet their capabilities. Even more importantly, artists' work doesn't monotonically get better. This idea that you're going to climb levels in this sort of strictly increasing way, I think is deeply inauthentic. Some of your works are going to be hits, and then you're going to go through a lull. Your work might not be great. The community might not appreciate it. Then your next piece is going to be a masterpiece. How do you get into a place where you can be committed to your talent over a long duration? There's going to be ups and downs, but they're so fundamentally talented that you're going to stick with them. You're not going to be judging them work to work. It's not a daily mark to market. It's like, look, you are special. This is a home for you, and we value you over this long arc of what we're building together. I think there's an element of really great talent spotters just know. I don't think I'm the best at talent spotting. I think I'm okay at it. But when I think about Dr. Karp, when I think about Joe, the number of the first 50 people that Joe personally recruited and talent spotted when they had no PR, they were 22, 23, there's no track record. It's not like you're looking at a prior body of work and saying, look at what you've done. It's just like, I still kind of look back at how Joe picked those people, and it's like, how? It's incredible. I think there's a real skill to that, and you have that ability. You're almost like a talent agent in the Hollywood sense. You're the modern Ovitz of Silicon Valley. How do you manage talent? It's a very open-ended question, but I think that's roughly how you would think about it. It's like, look, I'm really here to support. I need to create the environment that allows your art to flourish. I'm not the one managing your art. If you think about that, they need a huge amount of autonomy. They need the resources. They need some protections from the daily distractions of certain aspects of the business. I think if you're always inhaling the full visceral scary reality of everything going on, I think that distracts you from doing the art. How do you have complete autonomy over an area that you're very, very good at, and the ability then to execute against it? I think a kind of maximum I always think about is smart people cannot be taught anything; they can learn. That's the environment you're trying to set up. It's like, how do I create this environment of maximum learning? It's like, would I be excited to stand shoulder to shoulder with this person over the next decade? Can I imagine continuing to invest and grow in them to take on more problems? If they're talented, I'll figure out the role. They'll figure out the role. We have so many needs.

When you've been wrong, the less obvious part of this sort of reinvention is reinventing yourself. It's the toll on the individual where you're like, look, I got this thing, and it's at this place, and it's successful. Where is my earned credibility from having done that? In the reinvention, what you're saying is, no, no, we're all, every single person, we're starting from zero. It might be true that some person who joined this company two months ago is going to be more sophisticated at this new problem than you who joined five years ago. There's a commitment to pushing through that. I think it's actually only survivable because the whole thing is set up as an artist colony. Are you not screwed if I was a model company? I would be pushing into the app layer. I'd be trying to figure out how to own more of the value I'm creating with the model rather than just selling an endpoint. This is usually the opposite advice you're going to get from traditional venture, where they're like, build something simple that scales with an API, because otherwise you're completely constrained to the parametric knowledge of the model. I think that's a Sisyphean task. People keep trying to jam more and more knowledge into the model, but what they fail to recognize is that these models are actually very good at speaking regular grammars: JSON, code. It's a regular grammar. How do you use these models in their current state to manage and manipulate application state? I think the world is likely to change much more in the next 5 years than most people are modeling. The geopolitics are highly unstable. That maybe is a more obvious statement now to say after October 7th. I think we're entering a very challenging time here, and I think that has consequences for the commercial world as well as for governments. The other aspect of this is, as we approach that sort of instability, what is the role for real transfer? We have this unique opportunity with AI to actually change things, not just to build businesses, but to fix many of the problems that drive at the fundamental legitimacy of the institutions we have in the world. Why don't these institutions work well? I am not someone who thinks you're going to fix them by tearing them down. You're going to fix them by partnering with them and helping them, helping the humans who wake up every day trying to make their institutions work, work better.