Palantir 如何打造终极创始人工厂 | Nabeel S. Qureshi(创始人、作家、前 Palantir 员工)
How Palantir built the ultimate founder factory | Nabeel S. Qureshi (founder, writer, ex-Palantir)
Palantir produces top product leaders
Lenny Rachitsky: 30% of PMs that leave Palantir start a company. Just give us a picture of what the people are like.
Nabeel S. Qureshi: I feel like they screened really hard for a few traits in particular. One is like very independent-minded people who weren’t afraid to push back. Two is people with broader intellectual interests.
Nabeel’s Palantir experience and PM stats
Lenny Rachitsky: What’s the difference between, say, a PM at Palantir versus a traditional PM?
Palantir’s talent filtering and culture
Nabeel S. Qureshi: They were extremely careful about only making people PMs who had first proven themselves out as forward deployed engineers. You basically could not become a PM any other way. There’s two types of engineer at Palantir. So, there’s one that works on the core product and they’re a traditional software engineer. There was a different type of engineer which you sent into the field. You would spend maybe Monday to Thursday and you would actually go into the building where the customer worked and you would work alongside them. You would literally get a desk there and so, that engineer became known as a forward deployed engineer.
Lenny Rachitsky: What’s something that you believe that most other people don’t?
Red flags and mission appeal
Nabeel S. Qureshi: I think this is a somewhat contrarian view within tech.
Lenny Rachitsky: Today, my guest is Nabeel Qureshi. Nabeel is a founder, a writer, a researcher, and an engineer. He was recently a visiting scholar researching AI policy at the Mercatus Center alongside Tyler Cowen. At one point, he worked with the National Institute of Health and major clinical centers to create the largest medical data set in the world. He worked at the Bank of England for a bit. He was founding member and VP of Business Development at GoCardless, one of Europe’s biggest financial technology unicorns.
And most related to the topic of this conversation, Nabeel spent almost eight years at Palantir as a forward deployed engineer working on public health projects with US federal agencies, including public health services during the COVID-19 response and applied AI in drug discovery. Whether you are a fan of Palantir or hate everything that they do, they are an important and fast-growing company that is pumping out incredible product leaders, as you’ll hear more than any other company in the world. So, it is worth studying and understanding.
I’ve never heard an in-depth conversation digging into how they operate, build product, hire, and were able to scale from a primarily services business to a software business. So, I am very excited to bring you this inside look. In our conversation, we go deep into what the heck does Palantir even do, why getting good at managing lots of data is an underappreciated secret to their success, a look at the unique forward deployed engineer role that they innovated, and what other companies can borrow from their insights here. Also, how they hire and how they build amazing product leaders, plus a ton of advice on talking to customers, building products, and starting companies.
If you enjoy this podcast, don’t forget to subscribe and follow it in your favorite podcasting app or YouTube. Also, if you become an annual subscriber of my newsletter, you get a bunch of amazing products for free for a year, including Superhuman, Notion, Linear, Perplexity, Granola and more. Check it out at lennysnewsletter.com and click Bundle.
With that, I bring you Nabeel Qureshi.
WorkOS also recently acquired Warrant, the fine-grained authorization service. Warrant’s product is based on a groundbreaking authorization system called Zanzibar, which was originally designed for Google to power Google Docs and YouTube. This enables fast authorization checks at enormous scale while maintaining a flexible model that can be adapted to even the most complex use cases.
If you’re currently looking to build role-based access control or other enterprise features like single sign-on, SCIM or user management, you should consider WorkOS. It’s a drop-in replacement for auth zero and supports up to one million monthly active users for free. Check it out at WorkOS.com to learn more. That’s WorkOS.com.
With Attio, AI isn’t just a feature, it’s the foundation. You can do things like instantly prospect and route leads with research agents, get real-time insights from AI using customer conversations, and build powerful AI automations for your most complex workflows. Industry leaders like Flatfile, Replicate and Modal are already experiencing what’s next for CRM. Go to attio.com/lenny to get 15% off your first year. That’s attio.com/lenny.
Nabeel, thank you so much for being here. Welcome to the podcast.
The “Inquisition Board” mechanism
Nabeel S. Qureshi: Thanks, Lenny. Glad to be here.
Why competitiveness finds the right fit
Lenny Rachitsky: In our chat today, I want to zero in on a post that you recently wrote where you shared your reflections on your time at Palantir. You spent something, maybe just under eight years there. The reason I’m really interested in Palantir is I’ve been doing a bunch of research recently looking into which companies hire the best product managers and create the best product managers, and Palantir just keeps coming up over and over in the work that I’m doing.
So, I’ll share a few stats real quick. I looked at which companies produce the most founders, especially out of their PM team, and Palantir is, by far, number one. 30% of PMs that leave Palantir start a company. And number two is 18% and that’s Intercom. So, that stat, I looked at which companies PMs that leave get immediately promoted in their next role, Palantir is number one of all companies in the world.
I looked at which companies’ PMs become the first PM at another startup that they join, Palantir is number two in the world. And then I looked at which companies alumni PMs become heads of Product down later in their career, Palantir is number three in the world. Also, just the company is doing extremely well. It’s worth, I think, something like $200 billion these days. So, there’s a lot to learn from Palantir.
I actually want to start a question that I imagine every employee at Palantir constantly gets that, and I still don’t think people totally have an answer in their head. What does Palantir do?
Nabeel S. Qureshi: That’s a great question. You started off with an easy one, Lenny. So, Palantir is, the way I describe it, is they achieve outcomes for their customers very tactically. The way they do that tends to be through a data platform. So, they have what I consider to be the world’s best data platform, and I can go into what that means in a second. And then there’s a couple of different versions of this. So, there’s one that’s optimized for intelligence and defense use cases that one is called Gotham. And then there’s one that’s more optimized for commercial use cases and that one’s called Foundry.
And that’s the classic explanation of what they do. So, they sell a data platform. They typically work with very large customers is the other thing. So, it’s going to be Fortune 50. It’s going to be governments around the world. It’s going to be those kinds of customers. So, that’s the capsule answer, but there’s lots to unpack in there.
Why Palantir is a founder bootcamp
Lenny Rachitsky: Awesome. Okay, and we’re going to touch on a lot of this stuff, including the data piece. I want to start with talking about just the people and the culture of Palantir. You shared a bunch of really funny stories of what it’s like to come to work and even interview at Palantir. There’s a story you shared where because maybe the co-founder, you’re walking by and he’s chewing ice, and that’s some benefits to cognition. Just give us a picture of what the people are like, especially early days Palantir and the culture and how unique it might seem.
The no-title company
Nabeel S. Qureshi: Yeah, it’s definitely, it’s an add-a-one company. I don’t know how else you would start this company if you were not somebody like Peter Thiel. And so far as, it seems like there was a point at which they owned a silly fraction of the office space in Palo Alto. So, you’d walk around Palo Alto and there would just be Palantir hoodies, Palantir buildings everywhere and so on.
And so, I feel like what happened at some point is they raised a lot of money and they resorted to all these really interesting ways of just getting top talent out of places like Stanford and other top schools and just people who knew the founders who tended to be very interesting intellectual people. And I feel like they screened really hard for a few traits in particular. So, I would say one is very independent-minded people, people who weren’t afraid to push back, who questioned the frame of everything and thought for themselves and had strong convictions.
Two is just people with broader intellectual interests. Karp just released a new book and he’s quoting Habermas and all these European intellectuals and just things you don’t typically see a tech CEO do. And so, I think there’s that intellectual strand in the company. And then yeah, I think three is just people who are very intensely competitive. There’s a sort of win at all costs mentality to the company. And so, I think those were the set of traits that were this gravity while in California at a certain time. And so, you just had a lot of really fascinating people joining the company at that time.
The way they screened for this was interesting too. So, for the longest time, they had… Everyone does this now, I think, but at the time, it was a little bit rarer, is a founder had to interview you in order for you to receive an offer. And so, a founder, it could have been Alex Karp, it could have been Stephen Cohen. Earlier on, it might have been somebody like Joe Lonsdale, but it was always one of these people.
And the interviews were pretty strange. With Stephen, it would be, you’d be chatting about philosophy for an hour and a half and it would very much just be like he would pick a topic out of thin air. It was impossible to prepare for, and then he would just go very, very deep and try and test the limits of your understanding. But it would really just be a fun conversation and then if you pass the vibe check, you’d be in. And so, there was that strong selection mechanism.
There was also the question of, I think it might have been Thiel who mentioned this, but he thinks that a lot of the best recruiters in the world or the companies that attract talent, they put out this distinctive bad signal and it has to turn some people off. That’s the key of a good, bad signal. So, I think in the present day, OpenAI and Anthropic, they’re both sucking up some of the best talent that you and I know. And I think one way they do do that, and they are sincere in this, but they do really attract people who are almost messianic about the potential of artificial super intelligence and who really believe this is the only thing that matters and it is going to be the biggest thing in the world.
I think Palantir’s version of that was that they were quite focused on things like preserving the West. There was a slogan of Save the Shire, right? So, they were talking about military and defense and intelligence and the importance of that well before everybody else. And bear in mind, this was during the era when it was social, mobile, local apps. You had, social media was on the rise. You had, the hot companies were Facebook and Pinterest and things like that. And so, this was, at the time, a very strange thing.
And so, I think to be drawn to that, you had to look at the other options and say, “Well, this is fine, but what am I really doing in life?” Whereas you had this other place that was like, “Hey, come solve the hardest, messiest problems in the world with us.” And I think just at that time, that really drew some really good people.
What is a Forward Deployed Engineer
Lenny Rachitsky: We’re going to talk about the reasons people don’t necessarily like Palantir and the moral question of what they do, but when people look at a company that is like… I guess OpenAI, to your point, is a good example where they’re just so turned off by maybe their approach. What you’re missing is that’s potentially intentional because it actually draws in the people they really want.
It makes me think about, I was involved in creating the core values at Airbnb and something that we learned at going through that process is, when you define the values for your company, it’s really important to clarify who this is not for, exactly as you described, which feels unnatural. Like, “Oh, we want to be inclusive, we don’t want to make people feel like they don’t belong.” But the whole idea is to be clear on here’s who will thrive here and here’s who’s aligned with our mission. And what I’m hearing is Palantir and these companies take it to the extreme.
Two types of Forward Deployed Engineers
Nabeel S. Qureshi: A hundred percent, yeah. On my team at Palantir, one process that we followed, I could talk about this more if it’s interesting, is when you started a new project, you basically had to organize what they called a murder board for it. I think this is originally an army type. So, the idea is, basically, you write up a two-page plan for the project. You invite three or four smart folks who don’t know anything about the project and their job is just to tear apart your plan.
And so, you have to write, here’s the vision for this, here are the goals, here are the tactics over the next three months. And one section was principles that you’re following for this project. And I remember giving this advice a lot was just like when people joined, they would write principles such as move fast and I would always be, “Everyone likes to move fast.” It is not a good principle actually because nobody can really disagree with this reasonably. You need something that actually a lot of people are going to go, “Why are you taking this principle? This seems wrong to me.” So, you need something that people can disagree with.
Lenny Rachitsky: I want to come back to the beginning of what you described of what they look for, what Palantir looks for in people. You talked about independent-minded, a lot of interests, broad interests, and competitive. First of all, I think a lot of people hearing that, especially the last part, be like, “I don’t want to work there.” Why does this work? Because this isn’t naturally what you would think of as how you build the most amazing, productive team.
Inside an FDE’s daily work
Nabeel S. Qureshi: I think it just draws people who want to win. I think that’s what was really important. The other piece of it, I think, is that there’s actually, and this was much truer 10 years ago, is there was a lot of talent that was a little bit outside of the tech ecosystem but could easily have been very successful within it. So, people who got out of the military or one of the intelligence agencies and they were doing, let’s say, an MBA somewhere to transition into the corporate world. And I think, typically, they would have taken a position at a classic Fortune 500 corporation. And actually, Palantir managed to get a bunch of that talent. And at the time, that was very undervalued.
The people who succeed the most in the Marines or the Special Forces or whatever it is, tend to be pretty smart people. They tend to have accomplished very difficult goals in very hostile environments. And it turns out that when you’re starting a somewhat chaotic tech company, that’s actually a very useful skill to have. Again, I think more companies are doing this now, so Scale AI and et cetera. But at the time, that was a very differentiated talent pool.
And so, I think having those values as opposed to maybe the values that were more in fashion then, so talking about how inclusive you are, or the sushi that you serve at lunch, or whatever it is, it just drew a very different crowd. And I think the game that was being played there was, one, it’s mission alignment. You’re doing a defense company, that’s the kind of person you want to attract. But I think there’s also, two, which is just what is the talent that maybe is a little bit undervalued now and how do you actually draw those people to you? And I think that game is always shifting.
Lenny Rachitsky: This is definitely starting to explain why so many Palantir alumni go on to start companies and become leaders at other companies. These are leaders that you’re hiring. So, it feels like a lot of it is just the talent you hire are people that are naturally leaders.
Value-based pricing business model
Nabeel S. Qureshi: I think you’re right, and we can get more into it, but I think there was also a very concrete set of ways where that place was a training ground for founders. I even think it turned a lot of people who might not have become founders into good founders because of the way it works. So, I think there was a selection effect there, but there is also some training effect too, but it’s unique to the way the company works.
Lenny Rachitsky: And is that along the lines of the forward deployed engineer stuff or is that something else?
The problem-to-product loop
Nabeel S. Qureshi: It is that, yes.
Weekly rapid iteration cadence
Lenny Rachitsky: Okay, cool. We’re going to get to that. I love it. Okay, amazing. Before we do that, one last thing is something I’ve seen is that you guys at Palantir don’t really have titles. Everyone’s the same level and just generic titles for everyone. Talk about that. Why do you think that was important? Why was that useful?
From services to platform
Nabeel S. Qureshi: I don’t know this for sure, but I do know that Thiel writes about this in Zero to One and his take is just that as soon as we have these title, you have a thing that people are competing for and then you get these very unproductive conflicts. You get people optimizing to game the system. You get Goodhart’s law everywhere. So, it’s like you have a metric and then people basically manage to the metrics.
I don’t want to pick on any one company, but if you take Google, for example, there’s a lot of interesting posts by people who left Google and they cite this as a reason why they got a little bit disgruntled, is that there’s a way to get promoted. Rather than, let’s say, improving an existing product, what you do is you start a completely new product and that has your name attached to it. And then when it comes to promotion season, you could say, “Hey, I did this new thing.” And then boom, you have a new Google product, but it’s maybe confusing to the end user.
So, I think they wanted to avoid all these kinds of dynamics. And so, the way that they did that was they said, “Well, titles are not going to be this memetic totem that everybody competes for. Instead, everyone is just going to have the same slightly meaningless title, which is forward deployed engineer.” And the only people who did have titles were the CEO and then there were six directors and that was it. And now, I think it’s a little bit more nuanced. There are different teams. There are some people with titles, but honestly, it was almost like…
We used to joke about it. It’s like people would leave the company and then you’d see them update their LinkedIn and they would be like, “Oh yeah, I was totally the SVP of XYZ.” And it’s like, “No, you weren’t. You’re just…” But then it’s like I totally understand it too because when you leave the company, you have to make your experience legible to the next person. And so, guess what? Things like SVP actually do matter.
And so, yeah, I think they wanted to avoid this intel competition. There are downsides to doing this. So, maybe the competition isn’t as explicit around a specific title, but instead, what it becomes about is there’s a particular exact or something and you want to gain that favor. And so, it becomes more about who can get in the inner circle of this person or whatever. And there were those dynamics too.
I actually am a big fan of this philosophy though, the no titles one. I think what it did do is that it basically said if you are in, let’s say you’re in a role of you’re leading a very important project, which could happen, what it said was… This is always fluid. So, you are in this role because you’re very good and so, it’s a meritocratic thing. But if you start performing well, it’s actually very easy to shift that because there is no explicit ” I am the GM of this project title.” And so, you always had to earn your place in the company. You always had to earn the right to work on what you were working on, and I think that was a good side effect.
Foundry vs Gotham differences
Lenny Rachitsky: Let’s start talking about forward deployed engineers. What is a forward deployed engineer?
Nabeel S. Qureshi: So, the way this originated was, basically, you can think of it as there’s two types of engineer at Palantir. So, there’s one that works on the core products. So, they don’t necessarily leave the building in Palo Alto or New York or the offices. They’re very much working on the core products and they’re a traditional software engineer.
Because of the way the company works where you had these very large engagements with these large entities, there was a different type of engineer which you sent into the field. So, what that meant was you would spend maybe Monday to Thursday and you would actually go into the building where the customer worked and you would work alongside them. You would literally get a desk there. And so, that engineer became known as a forward deployed engineer.
So, within the company, that function is known as business development or BD. And then PD is product development. So, it’s where the product is made. And so, within BD, you had forward deployed engineers. There are actually two types. So, there is one that it’s a more technical software engineer. So, you have to pass a software engineering interview and prove your chops there and you would typically have a CS degree, but there was actually a type of forward deployed engineer that didn’t have that. So, you would still get a technical interview, but it would be less about, do you know the specifics of this C++ algorithm? And it would be more about just like can you reason about data? We didn’t have that division originally, but it turns out that there’s a lot of people who are technical adjacent, shall we say, who you really need in the room when you’re working with these large organizations or these large companies, because translating what you’re doing into language that would resonate with an executive or being able to navigate the social dynamics in a room, all these are very valuable skills. And so, the hiring criteria there were a little different. It was a bit more about, are you savvy as a human? But all of that was given the title of forward deployed engineer, and it’s just an engineer who works with customers.
The Airbus project story
Lenny Rachitsky: Okay, so just to make this crystal clear for people, a lot of people hear this idea of Palantir having forward deployed engineers. A few other companies have done this. It’s pretty radical. So, as you described, you basically have a desk at a company. So, you worked with Airbus and we’ll talk about that. So, let’s just make it real. So, you have a desk and a computer and login access and all these things at Airbus. You go to their office four times a week. You’re sitting there with their employees working side by side, building a product for them, versus what most people do where “they just talk to customers,” where they do an interview once in a while, they do a Zoom, they share mocks, things like that. This is like that on steroids. Is that roughly the way to think about it?
Building factory management tools for Airbus
Nabeel S. Qureshi: It is, yeah. And so, we would really be there a lot of the time. And so, the side effect of that was, one, you learn to live and breathe the customer’s problems and you learn to speak their language. And eventually, they saw you as one of them and so, you develop these really close bonds with the customers. So, at Airbus, I would be at the factory where the planes were produced, or I’d be sitting next to people diagnosing issues with aircraft or whatever it was. Similarly, later on, I worked with the NIH, which was part of the US government, and I actually had a badge there and I would work with civil servants and biologists and clinicians and people who were working there.
And so, it’s this pretty radical thing as you suggest. I think the key thing there from a business point of view is the average deal that Palantir had was very large in the many, many millions of dollars, which means that you could pay for this as part of the thing that the customer got. And then it was priced according to the value that the customer got.
So, as a simple example, if you’re Airbus and let’s say that you have an issue with one of your planes and you need to fix it, and fixing that is worth a $100 million or something to you, that’s how it would be priced. It would not be priced as, “Hey, you’re buying data infrastructure and it’s similar to Snowflake or Databricks or one of these other providers. It’s much more anchored to, here is the outcome.
But then the job of the forward deployed engineer is not just to deploy software. It is not just to sell software. It is to actually solve the problem. And so, you would have to be there. You would have to meet the key stakeholders who are actually in charge of reporting to the CEO about the specific issue. You would have to become their friend. You would have to gain their trust. And you would have to, in some cases, create new software such that it could actually solve the novel problem that was in front of you.
So, I would have friends who worked with one of our energy company customers and they would have to learn the ins and outs of how oil wells work. And then out of that, it turns out that having streaming data is actually very valuable for this use case. And so, boom, suddenly, there’s a product that can handle streaming data that becomes part of the core platform. But that would be the motion, is you learn about the problem. You figure out what software would best address it. You build that software. You use it to accomplish the goal. And then eventually, that gets folded into the broader product suite.
And so, you can start to see why this would be a good forge for founders. And this was actually part of my thesis going in and joining, was I said, “Well, say, I got five reps of this,” which I got more than that. But say, you got five reps of doing this in five disparate contexts, you actually become very good at this cycle of, okay, go into the building, gain the trust of the person, meet the people that are going to become your users, talk to them about their problems, make sure you’re building something that actually solves them, and it’s just a boondoggle.
Get really fast feedback and iteration loops. So, every week, you would have a cadence where it’s like Monday, you go in. You do your meetings. Monday night, you build something. Tuesday, you show it to somebody. Tuesday, you get the feedback. Tuesday night, you iterate on it. Wednesday, you show it to somebody. Wednesday night, you iterate on it. So, you get four of these, five of these cycles every single week.
It already got it. So you get four of these, five of these cycles every single week, and you’re moving incredibly fast. So 6 weeks in, you’ve suddenly gotten to, wow, this is really valuable, and somebody’s willing to pay you whatever, $20 million for it, and boom. I think this is why you get so many founders coming out of this same process.
Lenny Rachitsky: It’s becoming very clear why so many founders emerged out of Ballinger. Okay. So an important element of this as you described, is that the idea here is build this as a one- off solution to solve a real problem at say Airbus or some government organization. And then the idea as you create something out of that, that then Ballinger can sell to other companies. What’s extra cool about that is they pay you to solve this problem for them and then that is funding this other product that Ballinger can now sell to everyone. What a cool business.
However, early days Ballinger, everyone thought it was just this services business or just consultants building software for companies like Airbus, there’s no way they can make this a platform that works for a lot of people. Clearly, that’s what’s happening and it worked out. This is like the holy grail. Solve one customer’s problem and then sell it to everyone else. Every SaaS business basically would love to do this. What do you think allowed them to actually achieve this and be good at this? What are some principles that worked?
SAP data and the Ontology birth
Nabeel S. Qureshi: Yeah. That’s a great question and it’s true. I think that from when I joined until maybe until IPO and a little bit after, I was told, “Hey, isn’t this basically a sparkling extension? Isn’t it a consulting business lopping as a product company?” And eventually it became undeniable. One, because I always laugh when people are like, “What does Palantir do?” It’s like you can go onto YouTube and just search Palantir demo and you’ll get plenty of demos of how the software looks. Not many people know about this, but you can go and sign up with a credit card right now and start using it.
The limits of the FDE model
Lenny Rachitsky: I can have a Palantir account?
Product vision vs customer needs
Nabeel S. Qureshi: You actually can. Yeah.
Key elements of a great FDE
Lenny Rachitsky: I did not know that. That’s cool.
Nabeel S. Qureshi: Yeah. I think it’s called AIP now. So it’s not actually that mystical and there is a product, and if you look at the margins, they show that. So they have 80% plus margins, which is not really what you would get if you were actually a consulting company. It would be closer to 20 or 30%. So then your question was, well, how did they actually achieve this? I think there was just incredible talent in the product development organization, really top tier, incredible talent. And it took some really, really smart people to take the set of internal tools that we were using at the time to create value of customers and then go, what is the unified version of this? Would this look like if this were a product? And out of that process that I saw came Foundry assume there was a similar process with Gotham a while back. But basically it’s like, the motion was that you would go in and early on you were basically armed with Jupyter Notebooks and some data integration stuff, but it was very primitive and you had to create value that way.
But we kept building tooling that was useful for forward deployed engineers. So we were our own first customers and at some point there was this concept of, “Wait, what if we take our internal tools and we let our customers use them?” And I remember at the time, this is a really radical idea. And then Shyam Sankar, I think he’s the CTO, maybe he’s the president now, he just mandated like, “Okay. Every customer deployment you have to have a customer using this within three months or whatever.” So it was horrible at the time because these had been built for these nerdy Silicon Valley engineers, and so they weren’t particularly usable. They would crash all the time. You’d have to debug spark errors or whatever it was. But basically that process brought a lot more rigor to our thinking about the product.
And out of that kind of, I would say three or four year process came the Foundry product. And then there was a lot of focus around things like performance and reliability and so on. That was all really painful. So yeah, I think the answer was just talent. And then there was this recognition that we do. We do know things that most people do not know about how data works in large organizations. That was the other thing. We discovered a lot of “secrets” in this process of living with customers for so long.
The basic one was just data integration is massively painful inside organizations. This is very hard to understand unless you’ve worked in a large organization, but it’s actually impossible to even now to get access to a lot of your own internal data that you need to do your job. So you’ll hear stories of people being like, “I’m trying to calculate our sales this quarter, and I had to wait six weeks for some other analytics team to get me this deliverable.” So just knowing problems like that and being able to focus our product efforts around those problems, meant that we were able to build something generalizable there.
Evolution of the Palantir FDE
Lenny Rachitsky: Okay. There’s a lot here. First of all, you talk about Gotham and Foundry. I know that we’ll link to videos of people checking these out, but just what’s the simplest way to understand what these two products do?
Nabeel S. Qureshi: So Gotham is optimized for military and defense use cases and intel as well. I would say they both have some things in common. So they both have, I would describe this almost as a pyramid where the bottom layer is data ingestion, the middle layer is data mapping, and then the top layer is anything that’s user facing. So any UI component. And then if you think of Foundry for a second, there’s different tools that allow you to ingest data to it. There’s different tools that allow you to easily build data pipelines and clean up data, which everybody has to do. And then there’s a bunch of tooling that allows you to build compelling UIs on top, do point and click analytics, do notebook style workflows, however technical you are. So that’s, I mean, when it’s a platform, it’s a suite of things that has a common data backing but contains a bunch of different applications.So I think that is somewhat true of Gotham as well. But when you log in, you see this unified interface.
So what is the actual difference then? I would say with Gotham, you’re looking much more at workflows like that involve maps, for example. So when you’re doing a military operation, a lot of the time you are going to be looking at a map and you are going to be monitoring the movement of troops or tanks or whatever it is. Another big difference is the idea of graph-based analysis. So Gotham, one of the use cases was finding combing through networks of terrorists and basically finding the bad guys. So being able to do queries that are graph-based was important. So it’s like, “Who is everybody that Lenny called in the last week?” Imagine all the nodes fanning out from there. And then it’s like, “Okay. Well, this one looks interesting. Let’s zoom in on that. What is this person’s location?”
So it’s this very graph-based way of thinking that also applies to things like fraud. So Gotham has been deployed against fraud, but if you look at Foundry, it doesn’t actually emphasize that component so much because it turns out, let’s say you’re a B2B SaaS company, you’re probably not doing that much graph-based analysis. You’re doing things that look a lot more like classic SQL queries, tables, that kind of stuff. So Foundry is a lot more traditional in that way.
The power of data
Lenny Rachitsky: That was an amazing explanation. For the first time, I am starting to understand what these products do. Basically, it’s just sucks in a bunch of data, cleans it up so you can actually trust it and then helps you interact with it in various use cases, maps, graphs, tables.
Nabeel S. Qureshi: Yes.
The data gatekeepers
Lenny Rachitsky: Okay. Amazing. The example you gave of what you worked on at Airbus, you described it as basically a sauna for making planes. Is that right?
Nabeel S. Qureshi: Yes.
Key hiring insights
Lenny Rachitsky: So how much of that does becomes a part of this core product versus stays this one-off thing? Is it elements, that’s a cool innovation, let’s put that into Foundry. How does that work?
Drive over skills in hiring
Nabeel S. Qureshi: This was a really interesting story actually. So the initial problem that we came into with Airbus was that they had a new aircraft called the A350 beautiful aircraft. By the way, if you get to, I think if you fly New York to Singapore, it’s often in that A350. Really nice. So it was a relatively new aircraft at the time, and their mandate to us was, “Okay. We need to ramp up production of this really fast,” much faster than we’ve ever done it before. So it’s like the numbers are very approximate, but it’s like, “Okay. We’re producing 4 this month, we need to do 8 the next month, 16 the month after, and so forth, and you are going to help us do it.” So this goes back to what I was saying earlier is the mandate wasn’t like, “Hey, we need to upgrade our data infrastructure. We thought you guys would be met the list of requirements.” It was much more just like, “Please help us accomplish this mission. This is the big thing.”
So we went in, scoped out the problem. There were a bunch of different things that we could build that helped accelerate this, but one of the basic problems that we figured out was that without getting too much into the weeds, the way the factory would work, is that there’s a bunch of stations and you can think of the plane as literally moving between each station and then each station would do a certain set of work on it. So initially, it’s literally a big fuselage and the fuselage is sitting there and then people are doing a bunch of work orders against it. They need parts in order to do that work. And then at some point they say, “Okay. This is ready to move to Station 31, and the plane is physically moved to the next station and then Station 31 does its next thing.”
So in order for the next station to do its work properly, they need to know, one, what work was done at the previous station and what work is remaining? Two is just like, if you think about this problem, not all work is going to get done on time. So things carry over to the next team, and the next team then has to… So when I’m describing this problem to you can start to visualize, okay, maybe I need some Gantt chart to this, and I need the ability to click in and say, “Okay. What did Station 30 do and what work orders remained undone?” And then it’s like, “Okay. For those work orders, what parts do I need and where in the factory might they be?” So this was very, very hard to do as it is. A lot of it was just relying on people going and having conversations with other people on the factory floor, and coming from tech where it’s maybe not as complicated as building aircraft, that is a phenomenally complicated process, but it is easy to see, okay, you can actually improve this problem with software.
All that data was stored in SAP and SAP is like established software. It’s good at what it does, but it’s not the most user-friendly necessarily, especially if you’re not an expert in how it stores data. The table names are very hard to understand and read. So one of the things we figured out was just if you can pull in these tables that may as well be written in completely alien language, the table name would just be like S3, F1_Z or something like that. And you’d have to know, okay, this is the table where the part ID is stored or something.
If you could pull in those tables and join them in the right ways, and then just map them to human concepts that humans can understand, so things like a part a work order, an aircraft, et cetera, and basically build a hierarchy or mapping between them, then what you can do is, a user can just log in and say, “Okay. Aircraft 79, where is that? Okay. It’s at Station 31. All right. These are the work orders, et cetera.” So you’ve translated it into a more human-legible thing.
So the thing we built, I slightly flippantly described it as Asana. It’s a little different. But basically that’s what it did, was it gave you a unified view of, okay, this is what’s going on inside the factory. This is the work that needs to be done on this particular plane. And then me today going to my job at Station 31, what work orders do I need to fulfill and where are the parts that I need to do that? So did this directly become a part of Foundry? Not exactly, because the way that other companies work is not going to be using this same set of concepts, but the overall idea of taking a bunch of tables, and then mapping them to human understandable concepts was a very powerful one.
So this actually resulted in a big piece of Foundry now, which they call Ontology. You’ve probably heard this term as you’ve seen… If you see Palantir presentations, they always talk about Ontology. This is what they actually mean by that, is it is a set of concepts that is understandable to you as a human and you are not having to go and dig around and do. You’re just able to say, “Where is the aircraft now and where is it going next?” So the ontology became a huge piece of Foundry. It was directly informed by the learnings that we had from building that application inside that factory. And I would say it’s still a very big differentiator today. I don’t think too many other companies ship this kind of stuff yet.
Product managers at Palantir
Lenny Rachitsky: Wow. I love how excited you still are about this. I could see it being so fulfilling to solve this big problem. I saw a stat that I think, 4X their productivity. What was the number there?
The PM training path
Nabeel S. Qureshi: Yeah. I don’t recall the exact stat, but we did ramp up production, I think at least 4X that 1 year, which I mean obviously, they did this and we just helped with it. But that CEO said that we played a critical part.
Palantir’s ethical controversies
Lenny Rachitsky: Also, you moved to France, I think for this. That was how forward deployed you were. You lived in France for how long?
Counterintuitive: the value of college
Nabeel S. Qureshi: Yeah. I lived in France for about a year and a half. The way they built their planes is they manufacture different components around Europe. So they build the tail in Spain and the fuselage in part of the UK and Germany and so forth. So they basically ship everything to France to be assembled at the end, which you can imagine this is a very messy process. So I was mostly in France, but there would be weeks where I’d have to fly between all these countries just to figure out where things were.
Lenny Rachitsky: In your post you wrote about how just the life of forward deployed engineers is pretty crazy. You just get a call sometimes like, “Hey, you’re flying to this random country tomorrow. Get ready.” Is that just life as a forward deployed engineer?
Nabeel S. Qureshi: It is. Yeah. The company had a very, I would say, aggressive attitude towards travel in the sense of when you join, you were basically told, “Look, you have to be okay with travel. Are you okay with that?” And the attitude, which again I think is a very founder friendly one is you need to be willing to just jump on a plane that night if that’s the best thing to do for this customer and if it’s going to get us to where it needs to be to win. So there were many times when it would be like, “I need to take this cross continental flight tomorrow for this particular thing because it will be useful.”
So I think that’s one of the takeaways for me was just being in person is so valuable when you are working with some external party, just going there for a few days and spending time with them, maybe going out for dinner. You build so much more trust than if you’re trying to close a customer over Zoom or do an engagement over Zoom. It’s just the vibe is completely different. So yeah, getting on a plane was a really cool part of our job for a very long time. This obviously changed around 2020 because COVID happened, the company IPO, and so there needed to be a bit more internal controls around this. But I would say pre-2020, this was a very big part of the culture.
Lenny Rachitsky: I’m excited to have Andrew Luo joining us today. Andrew is CEO of OneSchema, one of our longtime podcast sponsors. Welcome, Andrew.
Andrew Luo: Thanks for having me, Lenny. Great to be here.
Lenny Rachitsky: So what is new with OneSchema? I know that you work with some of my favorite companies like Ramp and Vanta and Watershed. I heard you guys launched a new data intake product that automates the hours of manual work that teams spent importing and mapping and integrating CSV and Excel files?
Andrew Luo: Yes. So we just launched the 2.0 of OneSchema FileFeeds. We have rebuilt it from the ground up with AI. We saw so many customers coming to us with teams of data engineers that struggled with the manual work required to clean messy spreadsheets. FileFeeds 2.0 allows non-technical teams to automate the process of transforming CSV and Excel files with just a simple prompt. We support all of the trickiest file integrations, SFTP, S3, and even email.
Lenny Rachitsky: I can tell you that if my team had to build integrations like this, how nice would it be to take this off our roadmap and instead use something like OneSchema?
Andrew Luo: Absolutely, Lenny. We’ve heard so many horror stories of outages from even just a single bad record in transactions, employee files, purchase orders, you name it. Debugging these issues is often finding a needle in a haystack. OneSchema stops any bad data from entering your system and automatically validates your files, generating error reports with the exact issues in all bad files.
Lenny Rachitsky: I know that importing incorrect data can cause all kinds of pain for your customers and quickly lose their trust. Andrew, thank you so much for joining me. If you want to learn more, head on over to oneschema.co. That’s oneschema.co.
There’s a lot of founders listening to this and a question that I’m thinking and they’re probably thinking, and there’s two questions here. One is how hardcore to go potentially with their own forward deployed operation. And then two is just how and a company I know is actually doing this, how far to go with one company’s problem and invest in just like we are going to nail solving this one customer’s problem with the hope that this is something we can abstract and sell as a big platform. So let me start there. And you’re building a company, any I guess insights or advice on just how far to go down this road of we’ll solve customer one’s problem and we bet that this is going to be a big opportunity for a lot of other companies?
Nabeel S. Qureshi: So I would say on the forward deployed piece, my friend Barry McCardel, the CEO of Hex, the analytics company, he wrote a really good post about this actually, and his take was just like, “You probably don’t need forward deployed engineers.” It’s very specific. But I think basically the thing there is you have to be willing to be quite almost wasteful. You have to be willing to invest a lot in finding the thing. And for that you just need a certain ticket size. So you need each customer’s revenue to be probably in the billions of dollars. If it’s below that, you’re probably not looking at a traditional forward deployed engineer motion. It’s something a little bit different.
So I think one thesis that a lot of people left Palantir with and started companies around was there’s a lot of customers that Palantir won’t serve because maybe they’re too small a ticket size. So actually you could go and do something like Palantir for those companies, but instead of charging them $5 million, you’re charging them 250K. So in a scenario like that, you might still have forward deployed engineers, but they’re not going to France and spending five days a week in a factory. It’s more like you’ll have one person and they’re looking after five different customer accounts. It’s more of that ratio in order to make the numbers work. So I think a lot of the principles can be abstracted from that experience, but it is a really specific sales motion that depends on a specific way of doing business.
I think to your other question, yeah, I think it’s obviously something that is very hard to give a general answer to. My main thing here is just that you can definitely tell when you are just doing consulting and when you are closer to building a product. And I think the error that people make more often than not is they are actually too stuck on their own product vision. That’s the mistake I’ve seen a little bit more actually than the other way around. If you go to an enterprise customer, and let’s say you think you’re doing analytics software and it turns out they don’t actually care about internal analytics this much, they actually have this other massive burning problem and they don’t have a good solution to it yet. I think a lot of people are unwilling to go and pivot to the big problem because they’re like, “Well, we’re analytics software and so maybe this customer is a fit for our thing,” and maybe that’s the right call. In some scenarios, that is the right call. You should go find a different customer where your thing resonates more.
In other scenarios, it’s actually the right call to pivot and just put everything on that big problem instead and then go and find other customers for that thing. There’s no hard and fast rule. I remember reading a really interesting post by, I think it was David Hsu from Retool who had this exact thing. I think he worked at Palantir for a while too. He said that they had the Retool product and it wasn’t getting any traction at all. And then he tried an outbound email campaign where he literally just changed the subject line to build internal tools easily. And then suddenly they started getting all these replies from CTOs who were just like, “Yeah. This is actually a huge pain point for me.”
But the exact same solution, they were previously framing it as, I think it was supercharged Excel or something like that, and nobody was biting. So they just changed the way they framed it and found a different set of buyers and succeeded that way. So yeah, no hard and fast rule, but I think it’s always you need to have this matrix of options in your mind and be very deliberate about which one you are going with and why.
Lenny Rachitsky: I think your piece of advice is really important there. Usually in your experience, you’re saying people index too far too? Like now, what they’re asking me to do is not what I think they need or what customers will need. You’re saying it’s actually more likely they’re right, and that’s maybe where you should be focusing more versus this abstract vision and original idea you had?
Nabeel S. Qureshi: I think so, yeah. I think it’s very hard to not be anchored to your own experience and your conceptions as a problem. And one thing I’ve seen in really strong founders is they’re able to drop a bunch of those assumptions and almost treat a new opportunity as a completely blank slate. And then just figure out how to reshape things so that you’re taking advantage of that, and that’s how you don’t get stuck at a local maximum.
Lenny Rachitsky: Your other piece of advice is also really great. So people hear this and they’re like, “We don’t afford an engineer to sit at one customer prospects office and build stuff for them.” But your point is you can have one for five different customers. They’re not there full time. They bounce around, but they’re… It’s almost like sales engineering, just like what you call it sparkling sales where they help make it successful. I know Looker is a famous example. They think they called them forward deployed engineers. Do you know any other companies by the way, that some version of forward deployed engineers?
Nabeel S. Qureshi: There’s a lot. I mean, I know that the AR-Labs are hiring forward deployed engineers now, they’re building forward deployed engineering teams and they could make it work, but I think there’s going to be key differences. I don’t see Anthropic going into an enterprise customer and building some entirely from scratch solution for them. It’s going to be something that leverages the Anthropic set of products. So there’s a lot of companies that have this label now, but I think what’s really confusing about, it’s just that it means a few different things. There’s another post by Ted Mabrey who’s I think the head of commercial at Palantir, and that’s a very good one too, to point with those too.
Lenny Rachitsky: So say someone was, “I want to try this sort of thing in my company,” what would be a few bullet points if things they should get right? You’re describing the spectrum of what people describe as forward deployed engineers, if they were to try to do this, what do you think they need to most do correctly for it to be successful?
Nabeel S. Qureshi: The key things that made our model work well, one, they were actually real engineers who could build product themselves. That’s a very big difference. I think a lot of the time companies will say, “This person’s a forward deployed engineer,” but actually they’re mostly there to be more of a solutions architect, or they’re not necessarily building anything to know, but they’re just listening and trying to find a way of deploying the existing product. They’re not empowered to do new product. So the really radical thing Palantir said was, “No. Go in and if you need a completely new product to do this, you can go ahead and build it.” And I think that’s really the key difference.
The other stuff I’ve already mentioned, the value of being in person, and I think building close personal bonds with your customers. I do think the better founders do this anyway. They’re on texting terms with their buyers, they become friends with them outside of work, and they see them as humans who they’re trying to help. I think this is very motivating, gaining a really deep understanding of the business that your customers are in and knowing how those dynamics work. So a simple example might be, say hospitals in America. It’s very counterintuitive to think of a hospital as a business. People think of it as it’s a place where you get healthcare, but actually if you view it the way a COO or CMO views it, it’s going to look very, very different too.
As a very simple example, sorry, this is a little bit dark, but how restaurants want to turn over tables as fast as possible in order to maximize for the day? Hospitals actually want to do the same with patients. They would like to treat you and then get you out of a bed so they can free up the bed to get a new person in there. So that’s not super intuitive, unless you think hard about how the revenue for that hospital works. But then once you think about it, you’re like, “This has a bunch of problems associated with it.” And you start to go into really interesting…
… problems associated with it, and you start to go in really interesting directions.
Lenny Rachitsky: There’s just like the words and memes, and take you a long way working and understanding it.
Nabeel S. Qureshi: Yes.
Lenny Rachitsky: Okay, so essentially the things you want to get right, make sure it’s in person, make sure the person is technical, make sure they have a deep understanding of the business and the problems they’re having. The technical piece is interesting with AI tools these days, making everyone technical in some sense. You could argue this is going to become more common, people can just open up Cursor, Windsurf and just start adding features.
Nabeel S. Qureshi: I think this is a really interesting thesis you’ve just hit on, and I expect to see a lot more startups that take advantage of that insight.
Lenny Rachitsky: Basically it makes forward deploying engineers cheaper.
Nabeel S. Qureshi: Exactly.
Lenny Rachitsky: What is the current state of forward deploying engineers at Palantir? How much has it changed over the past few years? If you join now, is this still something you can do?
Nabeel S. Qureshi: Yeah, of course. I should obviously emphasize that one, I left the company in 2023, and so this is just my personal view, I don’t speak for them. I think that if you think about it, one of the metrics that the company had to measure its own success was essentially revenue per engineer, and so the more “product leverage” you had, the higher that number was. So if you had to throw a lot of people at every marginal problem, then you weren’t doing so well at that because you’re basically building a new thing every single time, and you are in effect a consulting business. If on the other hand, every time you encounter a new customer, the product turns out to be relevant to them, then great, and so this product leverage metric was actually a very unique thing and kind of a North Star for the company for the whole time I was there.
If you reason that out, what that means is that in the early stage of the company, you will have a customer and then you might have five to 10 engineers working at that customer. And so over time you want that ratio to change. So you want it to be each customer, because the product is so powerful, maybe AI coding’s gotten a lot better, each customer you only need two people, and then maybe you actually get to a point where you can have one person looking after multiple customers. And I think that’s how the job has changed, is now it’s a little bit more about you have multiple customers, maybe you’re spending less deep time with each individual one of them, but it’s a lot clearer what problem you’re solving across multiple customers and you have more of a kind of defined offering.
And so I do think that has been a bit of a change, but the company remains a very interesting and dynamic place to be. In some sense the story’s only starting, because one lens through which you can view this company is they spent 20 years basically building the mother of all data foundations for every important institution in the world, and I guess what’s very valuable now that AI models are out is proprietary data that isn’t public. Suddenly you have access to that and you are in a very privileged position to help your customers deploy AI in a way that makes them successful, and that solves real business problems. That is essentially the bull thesis for this company and why it’s probably going to 100X again. And so it’s still a really interesting time to join but I do think the nature of the ratio of people to a customer, for example, is one big difference now.
Lenny Rachitsky: Not investment advice, but it might 100X. I totally understand why that might happen. So let’s talk about the data piece, you said that this was one of the secrets of Palantir’s success, this early insight into the power of ingesting data, cleaning data, being able to analyze and work with it. What a marketing share there, just what they figured out about why this is so valuable, why it’s so hard and how they achieved it?
Nabeel S. Qureshi: I think it’s just very obvious as soon as you step into a corporation and spend a couple of days there really, is you’re like, all right, let’s suppose your job is to increase sales, so the first thing you want to do is get a clear picture of what’s going on. All right, so let me go and query the sales database. Oh wait, where’s the sales database? I can’t get access to this. Okay, I need to file an access ticket. All right, now I have to wait one week. And so everywhere we went, this was the big pain point, was we have to wait six to eight weeks just to get data access, and then when you do get data access, it’s not like the data’s in an easily queryable format, you actually really have to know what you’re doing in order to get the right metrics out, and so on and so forth.
And so it turned out like, okay, it’s this iceberg analogy where the actual analysis is actually just the tip of the iceberg, it’s kind of the last five or 10%, and the 95% before that is, I am gaining access to the data, I am cleaning the data, I’m joining the data, I’m normalizing it, putting it all in the same format. And so once we spotted that, then it’s like, okay, there’s actually a lot of product to be built there just to make that process easier. People don’t think of Palantir as this place where innovative new product and UX ideas come out, but I actually think it’s been one of the most generative companies for that specifically in the last 20 years, it’s just that most of that didn’t see the light of day and so people don’t know. But if you look at the product primitives that they developed in order to make the things I just mentioned a lot easier, they’re actually really valuable and interesting and could probably form the basis of independent companies themselves.
And so, yeah, it just took every single step of that process became much, much easier once there was a software solution around it. So if you talk about data ingestion, there’s essentially a universal data adapter that’s part of Foundry. It can read anything, so JDBC, S3 buckets, whatever you want. It allows us to look into the data, maybe preview the first 20 rows, and then it allows you when you’re ready to set up a schedule and just pull it in on some cadence. That process alone for an engineer used to take a long time, especially pre-Vibe coding, and managing all those cron jobs and doing this analytics, VM somewhere inside the customer’s tenant was a huge pain.
And so you productize that piece, then it’s like, okay, once you have the data, it’s like how do you actually join it? What if you’re non-technical? Is there a way for a non-technical user to be able to join tables and see what the result is? And so there’s all these very fascinating business problems that, because I think the access was very difficult to get, and people hadn’t really solved before, and so there was a lot of white space to do some product innovation. So now I would say Foundry’s definitely the best data platform in the world just because it has all these different applications within it that solve these discrete parts. And it came out of this, years of painful experience, watching people have to clean data and join it and figure out what this table name meant and so on and so forth.
Lenny Rachitsky: You shared in your post this kind of evocative story of some people’s jobs is just to gate keep the data. They’re there to give you access to this very valuable data within the organization, and how hard it is to get. That was a lot of this work, is just breaking through those political battles of like, “Okay, we need this data for the good of the company and took a lot of work.” I guess anything there you want to add?
Nabeel S. Qureshi: It is, yeah. It’s a huge pain, and there are good reasons for it. It’s not like folks are malicious here. If you’re IT or if you’re an InfoSec type person, then your goal is to prevent data breaches and to make sure that sensitive information doesn’t spread too wide. And so what’s the easiest way to do that? It’s to lock the data down, basically be a gatekeeper for access. I think where it got a little bit more interesting was where your skills are valuable and depend on you being the gatekeeper. So what I mean by that is let’s say I’m the only guy who understands the way the sales calculation pipeline works and I write the SQL for it. All the requests from business SMEs come to me, I have a big queue of them, it takes me weeks to get through this queue. I have a great job, I have great job security, and people depend on me.
And so now along comes this company and they’re like, “Hey, actually we want to make sales data available to everyone and we want to make it point and click.” Suddenly you’re like, “Hey, hang on, what am I going to do?” And so that’s where I think there was a lot of difficulty and I always say people are like, what accounts as competitors? I don’t think it’s the ones that you would think of necessarily. Palantir’s biggest competitor is a company rolling its own solution, and so the biggest difference would just be a CIO saying, “I’m going to build my own data infrastructure, I’m going to own it, it’s going to be on top of one of the hyperscalers, and we’re all just going to do our own analytics ourselves.” And what we came along with, which was quite disruptive to this model, was saying, “No, actually all your data is going to get ingested into this one platform and everybody in your company is going to use it.” The trade-off is it’s going to be really, really easy for everyone to do things. But as you can imagine, some people weren’t a huge fan of that model.
Lenny Rachitsky: It feels like Glean is the biggest competitor to Palantir after I hear this, do you know about that company?
Nabeel S. Qureshi: I do, yeah, Glean looks amazing from the outside. So many differences there, I can totally see why you would say this, but-
Lenny Rachitsky: Clearly a different use case but it feels like the reason they’ve been successful is they figured out a lot of this data ingestion, permissions, search stuff. I never thought of it that way.
Nabeel S. Qureshi: Yeah.
Lenny Rachitsky: Interesting. Okay, I want to talk about hiring, you talked a bit about this. You’re starting a company again, what are some of the key lessons you’ve learned from your time at Palantir when you are hiring people for your company? I don’t know if you’re actually hiring people yet, maybe when you may start hiring.
Nabeel S. Qureshi: Yeah, we have six people at the moment, so a really reasonably small team. I think with hiring, it’s funny, man, there’s so much hiring advice online and you read it and you’re like, “Yeah, this is super obvious.” And then when you live it, you’re suddenly like, “Aah, this is why people say this.” So a few simple examples are I think the thing that is really hard to find is somebody who really, really has a lot about doing the thing and will go that kind of extra 20%. I think when you hire out, especially not to pick on them, but I think if you hired a [inaudible 01:00:17] right, it’s like people want a 400K a year job, they would like to work a certain number of hours, they would like to ship some code and then go home, that’s basically the model that you get accustomed to even if you don’t intend to when you work at a big company.
And so if you hire out of that for a really small startup, it can be really challenging because a lot of your success as a startup depends on each individual person being like, “No, I’m really going to, I’m work this evening if that’s what it takes to get this thing working, and I’m not just going to check my boxes, I’m actually going to look towards what is the real outcome that this business is trying to achieve.” And everything I’m saying feels kind of obvious, but when you actually feel that difference between somebody who’s just checking the boxes and somebody who’s kind of an animal in this way, they’ll actually go and pursue and accomplish the end outcome, that difference is very, very big and it matters so much for your first 20 people. And there’s no science to finding these people. It’s not like you can just put somebody who cares about outcomes in your JD and then suddenly you’ll get all these people applying.
So then it’s like, okay, well how do you screen for that and how do you find those types of people? And so that’s where it gets really interesting. I think that’s where the mission alignment comes in, and so you do have to find people who, for what you are doing, have this extra maybe private reason to care about it a little bit more than the average person. So I think for Palantir, they did hire a lot of vets, for example, or maybe people who were a little bit more patriotic or pro-America than the average tech employee, and those people had an extra reason to Palantir and an extra reason to try that little bit harder. And so what I’m doing is a little bit more in the kind of medical and health space, and so I think people who have themselves had experiences with this system have maybe had relatives go through difficult experiences with things like cancer or whatever it is. They’re just that extra bit motivated to really care about the thing you’re trying to do and then work that little bit harder, and so I think aggressively filtering early on to things like mission fit, how much have you cared about stuff in the past, and what’s an example.
You ask questions like, what’s the hardest you’ve ever worked to get something done and why? And that does differentiate a lot of people, a lot of people don’t actually have a great answer to that. So I would say that’s been a really big learning, is it’s less about testing for the right skills, yes, that’s important, two it’s much more about just who has that extra 20%.
Lenny Rachitsky: That is really interesting, everything you’ve shared is essentially around motivation, and drive, and passion, and kind of just commitment to working on this intently, and it’s almost like a second thought of just like, oh, also they’re really smart and skilled at stuff. It feels like that’s just table stakes and this is actually what makes the difference in your experience.
Nabeel S. Qureshi: Yeah, I totally agree, and I think it’s different for every business. So I think if you’re in a space like B2B SaaS where maybe it’s a little harder to tell the story of like, oh, this is so mission-critical, whatever, there are other ways of getting at this thing. So for example, I know a lot of people, again, it’s a little played out now, but I know a lot of people who for sales teams, they will explicitly go for people who were professional athletes or played sports in college, and it’s like, okay, what does that test for? It’s like you are very, very disciplined, you’re very, very goals and numbers oriented and you’re willing to just work really, really hard. And so there’s all these kind of lateral ways of getting at these qualities that I think you just have to be intentional about as a founder. As a personal example, I’m a runner and so I actually love meeting fellow runners and I kind of joke like, “Oh, maybe I’ll go higher from run clubs or something like that.”
But it’s just same with I play a lot of chess, I love meeting chess players. I’m not necessarily saying that’s the right kind of hire for me, but I think having this thing of here are some traits that seem uncorrelated, but which actually give you good signal to this person’s personality, those are actually really important. The last thing I’ll say just as a funny illustration of that concept is I think Max Levchin tells the story of somebody interviewing at PayPal early on and he passed all the skill interviews and then it just got to the final round and he said something about liking to shoot hoops, like he liked to play basketball, and they were like instant reject. The vibe here was like if you’re not a mega Linux nerd, hardcore computer person, then we don’t want you here, even if you actually passed all the tests just because you like to shoot hoops. Now whether that was the right call or the wrong call, don’t know, but that’s an example of what I’m talking about.
Lenny Rachitsky: I think that’s a great echo back. People hearing this may be like, “What the hell? How dare they do that?” But this is exactly what you said at the beginning of our conversation, that like an approach to building a generational business is to be very clear about who this is not for, and that’s okay, it’s your company, not everyone needs to work there. And it’s almost saving them time because they might realize this isn’t for me, this isn’t the people I want to be around necessarily. So I think it’s important to see that side of it, is it’s your business, it’s important to be clear about who is a good fit for the company and who’s not. Speaking of that, let’s talk about product management for a bit. I know Palantir PMs are not traditional product managers. I imagine people have the title, Product Manager at Palantir, okay, so if so, as far as you understand what’s the difference between say a PM at Palantir versus a traditional PM say at a FANG company?
Nabeel S. Qureshi: Palantir was, as far as I remember, quite anti-PM for a while and eventually we did need them because we just got more serious about product testing.
Lenny Rachitsky: Classic story, classic story.
Nabeel S. Qureshi: A classic story.
Lenny Rachitsky: In many companies,
Nabeel S. Qureshi: The big difference or one big difference I noticed was that they were extremely careful about only making people PMs who had first proven themselves out as forward deploying engineers. You basically could not become a PM any other way. So as an example, when I mentioned earlier the thing that we built for the plane factory, the person who was managing that deployment, she later became the PM for ontology, and it was just because she’d kind of proven her method in the field. And the reason for that’s pretty simple, it’s going to be someone who understand how customers work and has that customer empathy, and it’s going to be someone who has this drive to get things done because that’s what BD selected for. I think the failure mode that they were very, very averse to in traditional PMs was this kind of Google Docs syndrome of like, okay, I’m going to write my product requirement documents, and I’m going to manage it in this very sort of sane, rational way I think, so the company was really rigorous about that.
And so basically PMs were almost always internal promotions and they always came from BD. I am not aware of a single case where we took somebody who was a PM at a place like Google, which produces many excellent PMs and hired them successfully into Palantir, it’s just a very different vibe. So I think that was one thing. This is maybe more of a classic PM trait, but you just had to be either an engineer yourself or extremely good at working with engineers, and the ones I saw who succeeded the most were just best friends with their engineering team. And the team would always just be like one, it was called the group pm and then it would be a lot of very, very good engineers. And basically the success or failure mode was just do the engineers and trust you? I mentioned before Palantir how is very almost disagreeable personalities, and so if you didn’t gain the trust of engineering team pretty fast, you didn’t last very long.
Lenny Rachitsky: I think we’ve cracked the question of why are Palantir PM’s so successful? First of all, the hiring bar is just basically hiring for leaders in a lot of different ways, to this, I don’t know, forge for founders where they’re working with a company solving a real problem, building a real product that makes money, and then those are the people that become the PMs at Palantir and then they go on to leave and that’s why 30% of them end up starting companies, I’m surprised it’s not higher, or become first PMs at other companies or heads of product.
Nabeel S. Qureshi: Yeah, absolutely, it’s crazy. I was part of a pretty small team within Palantir, I think it was 20 to 25 people when I joined, and I think at least six of them now are either unicorn or just pre-unicorn founders from that group of 25 people, which is actually a crazy ratio. And then a bunch more have become founders recently at an earlier stage, so yeah, there’s all these little pockets of excellence and it’s been really interesting to see. I think the other thing that’s driving that a little bit is when you leave, it’s just such an interesting company to work at that I think the retention numbers were actually very high for that company. People would often stay a lot longer than maybe the average Valley tenure. And so when you left, it was really this decision of just something very specific is pulling you and you want to kind of play the next level of the game, and so it was very unusual for someone to leave and then join maybe a more traditional tech company. It’s sort of like you’re either going to go become a founder or why would you leave when there’s so many interesting different things to work on? And I know that sounds a little culty, but that’s what everyone thinks.
Lenny Rachitsky: I could totally see that. A lot of people that left Airbnb have never found something more meaningful, it’s just hard, especially if you’re early. There’s a stat that I didn’t share that I think is really interesting, and when you look at YC founders and where they’ve come from, I think you maybe shared in your post that there’s more YC ex-Palantir founders than there are ex-Google founders in spite of Google being something like 50 times bigger sample size.
Nabeel S. Qureshi: Yeah, yeah.
Lenny Rachitsky: Let’s talk about the moral question of Palantir. A lot of people probably seeing the title of this episode, hearing this, will not be excited about Palantir being highlighted and promoted, a lot of people kind of disagree with what Palantir’s doing. Builds products that kill people in some ways, they work with governments they don’t agree with. I know you wrote a really insightful way of how you approach this question when you decided to work at Palantir and how you see people tackle with this, can you just talk about the framework that you landed on and how you thought about this yourself?
Nabeel S. Qureshi: Yeah, it’s a really interesting topic, it’s definitely very nuanced. I think what I was trying to say in that post was a couple of things. One was that there was a lot of upside there. So I worked on the US Covid response, I have friends who worked on Operation Warp Speed, and these are all things that I think saved a lot of lives, and I was pretty focused while I was working at NIH on cancer research. And so to me, these were just obviously good things and you couldn’t do them anywhere else, and so that was alone a reason to stay. The question I had in that post was, well, okay, there are definitely going to be other pieces of this that people object to. So during the 2016 to 2020 era, it became a pretty common thing to go into work in New York and you’d have people protesting outside your office or doing all kinds of things. And so there was this question of, well, is this okay? And I think the point I was trying to make was it’s rare that disengagement is the correct answer, and I think it’s more recognized now, but especially then it went a bit too far.
So the famous example here is Google kind of disengaging with a Pentagon AI project just because some people felt that working with the Pentagon was itself morally bad. I think that’s a way to sort of the left of what the median American would say, I think the median American would say it’s fine to work on defense stuff within reason and assuming you’re doing largely good things, and so there was just this kind of almost arbitrage there at some point of just hang on, it’s not like working on defense is inherently evil, it’s actually a pretty interesting thing. And then there’s this question of, well, would you rather be in the room and making this better or not? And so I’m struggling with how much I can share here, but as a simple example, if you’re doing even a workflow, which I think many people would not be super comfortable with, let’s say you’re targeting somebody for some kind of strike. If you compare the way it’s done now to maybe the way it was done in 2010, it’s going to be a lot more targeted, it’s going to be a lot more accurate, and so you’ve actually improved that process and reduced the chance of error. Maybe you should feel good about that, right? Now, that is a bullet many people are not willing to bite.
I didn’t work on the defense side of the company myself, but I think you have to be okay with these kinds of grade zones and actually actively thinking about what you are doing. And that doesn’t mean that it’s always the right thing to do to work in a defense company. Maybe we go into a very dark future and we start being the bad guys in some ways, and then it’s probably not a great idea to work at a defense company. So it’s a shifting landscape but I kind of felt pretty strongly that a lot of people in tech just didn’t want to think about this at all.
So you have engineers now who are working on optimizing short form videos for higher engagement, and you sort of want to say to them like, “Hey, are you thinking about what this is doing to the brains of young children?” Or “Have you seen an 11-year-old kind of scrolling something for five hours and do you think this is a good thing?” And I think people don’t want to think about this stuff too much. I’m not saying I know the answer, but there was almost this refusal to look at what tech was doing from a political lens for a very long time. It was just like, “Hey, let us play with our toys, let us sit in our little park, and don’t bother us, and we’re just going to build cool stuff and launch it.”
And 2025, we’re in a very, very different state of the world, tech is involved in politics now, and politics basically came to tech. There’s this famous image of Mark Zuckerberg, he’s sitting in Congress and he kind of looks very pale, and he’s like, “Why have they dragged me in here again?” But I think tech went through this journey of, oh, we’re suddenly becoming important now, oh, we’re really, really important now, oh, we better stop playing this game of politics. And so I think what I’m saying now is a lot more consensus than it was 10 years ago, but at the time, the feeling was just like, “Look what we are doing is political, so you better engage with that.”
Lenny Rachitsky: I think when this became really real for a lot of people is with the Ukraine War, the government’s running out of certain vehicles and ammunition, we’re just not able to produce it, and then we’re like, “Oh, thank God for a company like Anduril and all these other tech companies that are actually ahead and keeping us ahead.” I think the only reason the US is ahead of that…
And keeping us ahead. I think the only reason the US is ahead of China and the space race is because SpaceX just is one company that just has been doing this for a long time. So I think a lot of people have kind of realized, okay, maybe we need these things.
Nabeel S. Qureshi: Right. And I would make this argument as well, it’s like people are like, well, how can you feel good about working in defense? And it’s like, well, you’re not going to feel great if China invades Taiwan, actually, you’re not going to. I think you are probably also not going to like that outcome. So we do just live in this world where you do need to build up deterrents to these things and they better be good. So to me, it didn’t feel that difficult of a question. I think when you zoom into particular things, they can be very difficult questions and there have been a bunch of those in the last couple of years. But yeah, again, disengagement isn’t the answer.
Lenny Rachitsky: Yeah. And again, it’s not for everyone. I think that’s an important kind of theme through this conversation is some companies … like, to build … sometimes to build a generational, really successful company, you need to turn some people off because that’s what brings in the best talent oftentimes.
Okay. Just a few more questions. Kind of like stepping back a little bit. You’re building a company again. What are a few core pieces of advice that you’re bringing to your new startup that will inform how you build this company, from your experience at Palantir? We talked about a lot of stuff. Is there anything, I don’t know if there are three things that you think are like, “I’m definitely going to do these things this way because it worked really well at Palantir.”
Nabeel S. Qureshi: One thing is probably just really fast iteration cycles. So placing a lot of bets and then being really rigorous about just going through that cycle very soon. I have this [inaudible 01:16:40]principles, and one of the things on there is basically saying EOP successes goes up the more bets you make, and it’s sort of a function of how many bets you make and the probability of success of those individual bets, right? And so one easy way to almost guarantee that you’ll hit something is just to make a lot of bets and then just kind of cycle through them very quickly. Now, obviously this is difficult, often this question of, well, is this bet actually failing or are we quitting too soon, kind of thing. But that’s kind of one principle I take, is just test this thing very early. You know, like the classic “why feed, why see” thing is just when you take something to a customer, ask them to pay you a lot of money and [inaudible 01:17:22] then find a new problem. Don’t wait three weeks, which is what every founding team typically does because you don’t have that kind of time.
I do think the importance of just having a really tight, distinctive internal culture and building a strong feeling of trust within a team is really important. And kind of like you mentioned with Airbnb, and people definitely felt this at Palantir, there was this feeling of like, well, you worked here, you must be good. I trust you, and all of that. And I think it’s so important to create that and you kind of know that feeling. That’s what … like, people ask me, should I go work at place X or should I just go be a founder straightaway? I don’t know the answer for everyone, but I will say one of the benefits of working at a place like that is you just have all these internal benchmarks now for, okay, this is what this should feel like, and if it doesn’t feel like that, we’re off. And I can’t imagine not having those benchmarks and just kind of having to figure it all out.
So yeah, I think that thing, too, is just distinctive, internal, strong team culture. And then I think for me, think three is just working with a really messy part of the real world. So I kind of joked when I left, like, I am excited to just do pure software. I’m excited to, I don’t know, I want to build an ID or something and just not have a support email even and all of that. But it turned out, look, my comparative advantage in a lot of ways was the networks I’d built and the experience I’d had in engaging with the messy parts of the world. And they do need technology a lot, right?
There’s this horrifying thought I have sometimes of just like, maybe we’ll get ATI in the next two years and the healthcare sector will still be broken and it will still be impossible to afford rent in New York City and build houses and all these things. And that may well become true. And so I think it’s important to engage with those parts of the world too, even though they’re really, really challenging. And I think the really nice thing about LLMs is that actually, there’s so many workflows now that are accessible to you as a tech founder and people are somehow more open to working with tech companies than they ever were before. Selling into the sectors of the economy in 2015, incredibly hard. I think now post the ChatGPT moment, people are willing to give chances to small startups that they weren’t willing to do previously. As you mentioned earlier, the cost of doing things like forward-deployed engineering has fallen by maybe five to 10 x now at least. And so there’s a lot of new possibilities and I’m excited to engage with the best.
Lenny Rachitsky: Wow, that is some alpha right there that you’re finding, that some of these very large organizations are more open to working with startups, because classically, investors don’t want to invest in companies that are going after healthcare companies and governments and things like that. So it is really interesting actually to hear.
I’m going to mirror back the tips you just shared, and there’s actually a secondary tip that I think is the more interesting piece. So the first thing you’re taking away is iterate quickly, but I love your tip of ask for lots of money quickly, early, to see if it’s an actual idea that people will pay lots of money for. And if not, move on. I love that.
The other is build a very distinct culture, but the piece you share there that I love even more is this idea of knowing what a high bar looks like, knowing what awesome A plus people look like, and you need to work at a company like Palantir to actually see that. So the advice there I feel is just work at a company that is amazing, first, with the best talent, to understand what that should look like, plus you build a network of those folks. So I think that’s really interesting.
And then the other pieces of advice you’re pulling away is work on really hard, messy problems because that’s where the biggest opportunities are, and it’s sounding like this is the easiest time to actually do that. Amazing.
Okay. I’m going to take us to a recurring theme on this podcast called AI Corner. And what we do in AI Corner is we share some way that .. and this is you sharing … some way that you’ve found AI to be useful in your day- to-day, either in life or in work. Is there any way you found some tool … some AI tool useful that you can share?
Nabeel S. Qureshi: Oh my gosh, there are so many. I’ll give you a few examples. So I use Wispr Flow quite a bit. So this is the talk to your keyboard and it will transcribe for you app. Very good. It’s just great when you are iterating very quickly with an LLM and sometimes you have to do these paragraph-long prompts and it’s just easier to speak into them. So Wispr Flow, I like.
Lenny Rachitsky: Just to double down on that, you press a button and you start talking-
Nabeel S. Qureshi: Yeah.
Lenny Rachitsky: And it’s writing out what you’re saying. Cool. And there have been these products for a long time, Dragon Dictate and all these guys. Is the difference now these are just very, very good now at actually transcribing what you’re saying?
Nabeel S. Qureshi: I think that’s right. Yeah, they use a really good model and so it rarely makes mistakes even when I think it’s quite challenging. And then, yeah, the UX I think they just nailed. So that’s really good one.
I love Claude Code for developing. Even though I have my complaints about it, there’s something just very addictive about just telling it what to do. And it’s basically something that you run within the terminal of your computer, and so you just type Claude, it opens up the Claude interface. It’s very cute, it’s very beautifully designed, and you just tell it what to do. And it actually operates on the file system directly. So if you’re like, “Hey, create a bunch of these files,” that’ll just do it and you don’t need to go and muck around inside Finder yourself. And then it’ll do these really complicated pull requests and it’ll basically execute them quite well. So to me, this is a very exciting kind of preview of AI agents.
Lenny Rachitsky: That’s what I was going to ask. So this is essentially an AI agent engineer. I didn’t know that’s what Claude Code did. Very cool.
Nabeel S. Qureshi: Yeah. It’s sort of a guided agent, but yeah, it is really sweet. And then yeah, I’m just enjoying … you know, every week there’s a new, wonderful new thing to play with. Last seven days, I’ve been testing Gemini Pro 2.5. Excellent model. I don’t love Google’s UX sometimes, but I was playing with that. And I use LLMs every day for all kinds of things. The other day I was doing taxes and I needed to classify a bunch of transactions based on some metadata, and so I just wrote a script up really quickly and it did that. So.
Lenny Rachitsky: I love just the smile on your face as you’re describing all these AI tools. I think a lot of people are just like, holy shit, I’m just overwhelmed with all the things I need to be paying attention to. All these things I’m hearing, all these tools I got to try. And I love just this vibe of just like, this is incredible and so fun. We need more of that.
Okay. I’m going to take us to another recurring segment on the podcast. You’re going to get a double whammy. Contrarian Corner. So here’s the question. What’s something that you believe that most other people don’t?
Nabeel S. Qureshi: I think going to college is great. I think this is a somewhat contrarian view within tech, maybe not in the broader economy, but I often see people saying just like, oh, if you can just drop out when you’re 18 and just start working, why would you go to college? And I think this is completely wrong, but maybe it’s good advice for 5% of the population who probably would’ve been to your fellows anyway. But college is one of the few times when you can just make really, really deep friendships. You are in typically a nice campus. If you’re in North America, you get to spend all of your time just thinking and writing papers and reading books and hanging out with your friends.
And it’s actually very precious and it’s very hard to find that kind of time after you turn 21 because you got to pay your rent, you’ve got to work, you’ve got to do all this stuff. Let’s say you make a bunch of money, you take a career break, it’s still … all your friends are working and you always feel like there’s a ticking time or on top of your head or something.
So just taking those three or four years at the very beginning and going really deep on lots of different intellectual topics and being able to try different things and discover more about yourself. I’m a big college fan. I can’t comment on the ROI or whatever. I personally think the ROI is great, even though the fees are kind of high in the U.S., but that’s probably my kind of contrarian within tech view is don’t drop out of college unless you have a really good reason.
Lenny Rachitsky: It’s so funny that that is contrarian and it does sound contrarian. I had a great time in college here. Here. Okay. Is there anything else, Nabeel, that you wanted to share or leave listeners with before we get to our very exciting lightning round?
Nabeel S. Qureshi: No, I think it’s a really exciting time in the world. I think AI can be exhausting, but it does really just open up the possibility of building a better world in all these ways. And so I think just reassess what you’re doing every couple of months and make sure that it’s aligned with where I think AI is going and make sure that you are working on something that you feel has very high potential if it succeeds. And I think that’s more important than ever now just because the amount of leverage we have with technology is at the highest point in history.
Lenny Rachitsky: Let me double click on that real quick. So for people that want to do what you’re describing, what helps you understand where AI is heading and just kind of align with it, are there places of information and news you find useful? Is it just play with it kind of thing? What would you recommend?
Nabeel S. Qureshi: This is the big question. I use X a lot to keep on top of AI, so I would just recommend finding a good Twitter list and maybe following people off of that. There’s some good newsletters. I really like Latent Space, I know his X handle, it’s Swyx, S-W-Y-X. I can’t remember his actual name, but that one is very good and it’s pretty technical. I would recommend trying to stick to the more technical newsletters if possible. I think there’s a lot of philosophy about AI or AI policy type stuff, and I think that’s good if that’s your area, but it’s an area where it’s very easy to have a lot of takes on it. You’re not necessarily learning a lot by reading those.
But I think it’s just important to know what’s going on and make sure you are revisiting your own workflows as often as possible. And just making sure that the people who went here are going to be the kind of hybrid cyborgs who fuse with the AIs. This actually played out in chess, if I can take a slight detour, is the chess players who succeeded the most in the mid 2010s especially were the ones who were really early adopters of neural network based chess engines. So when DeepMind did that thing, there was very quickly an open source version of it called Leela, and you find basically the very top players like Magnus Carlson, Fabiano, they were the ones who kind of mind melded the most with Leela and learned how it played and then kind of started copying its moves.
And so I think just becoming a cyborg to the extent that you can. And then I think there’s this barbell thing of, it’s also important to just leave everything at go touch grass just for your own mental sanity.
Lenny Rachitsky: Excellent advice. And with that, Nabeel, we’ve reached our very exciting lightning round. Are you ready?
Nabeel S. Qureshi: Yes.
Lenny Rachitsky: Here we go. What are two or three books that you find yourself recommending most to other people?
Nabeel S. Qureshi: The first one that comes to mind is Impro by Keith Johnstone. This is actually … I wrote about it in that essay. It’s one of the books that [inaudible 01:28:36] used to send to people. I just think it’s a really interesting book. So nominally speaking, it’s about improvisational theater, which I believe this guy was a pioneer of. He was a British guy, Keith Johnstone, active between the ’60s and the 80s I think. And Impro is just this really interesting book about creativity and how social behavior works and basically just what he taught his improv students. It’s a very weird book. It’s full of these unbelievably strange ideas. There’s a lot of very tactical things he tells you to do in the first chapter, for example, just to break out of your own mental frameworks, really just wild stuff.
He’ll tell you to walk backwards while counting down from a hundred and think about some problem that you’re struggling with and there’s all these kind of odd things. But the number of ideas per page I’ve found on that book is extremely high. The concepts about how social interaction works and how things like status and so on play into your social behavior are super important. And they made every kind of fully deployed engineer read that for the simple reason that I think it just helps you kind of read people better and interact with them better and become more conscious of how you are coming across and just modulate that.
Lenny Rachitsky: What is the title again?
Nabeel S. Qureshi: Impro.
Lenny Rachitsky: Impro. Okay, cool. We’ll link into it in the show notes.
Nabeel S. Qureshi: Yeah, so Impro is number one. I think just to go a little more highbrow, maybe Shakespeare’s history plays, there’s a set of them called the Henriad, so like Henry IV, Henry V, Henry VI. I find most people don’t read these, so they’ll read Hamlet or Macbeth or whatever, but the Henry one is absolutely incredible. You don’t have to be interested in British monarchy or British history in order to enjoy them. They’re actually some of the most interesting and insightful books I’ve read about power and how power works and politics and what the sacrifices that you might have to make if you want to be a successful king in that case. But it transfers over.
I think it is worth thinking really hard about, I think especially in a world where everything is kind of organized around these prominent figures and personalities now. When you think about the current administration, you think Trump, Elon, or when you think about AI, you think of Sam Dario, right? And so I think it’s important to understand, how do you think about these personalities and yeah, the kind of game that they’re playing. And Henry is actually … the Henriad is an incredible kind of set of books around that.
They’re also easy to read, which sounds hilarious when I say it, but you can read a Shakespeare play in a day. They’re sort of … I don’t know, they’re like 50 pages long. It’s not that bad. You have to get used to the language. Yes. But I would recommend that for sure. I guess you asked for two to three. I love High Output Management by Andy Grove. I just think that’s a great business book, and people tend to read summaries of it on the internet more than they actually read the book. But the actual book has a lot of really interesting stories and explanations about … I think the most powerful thing about that book is actually how Andy Grove thinks, and less any of the specific tactics there. And I think you don’t get that unless you read how he came up with all these things.
Lenny Rachitsky: Your first two books were extremely out there versus what other people have recommended, and the third book was the most recommended book on this podcast. So I love that spectrum that we just went on. Perfect. Okay, next question. Do you have a favorite recent movie or TV show that you’ve just really enjoyed?
Nabeel S. Qureshi: The last movie I really loved was a Decision to Leave. It’s a Korean movie. It’s by the Director of Old Boy, which maybe some people have heard of. It’s a great movie. I think it was released a couple years ago and the basic premise is, there’s a detective who is investigating a woman who’s accused of killing her husband, and he gradually starts falling for her, which starts to affect his judgment in all these ways. Just a really fascinating kind of psychological thriller with a sort of romantic element to it. Visually, very beautiful. Yeah, I think a lot of the most interesting movies nowadays come from abroad actually. So East Asia, South Asia, places like that. TV, I don’t watch so much yet. It’s been a while.
Lenny Rachitsky: Totally understandable for a founder. Okay, next question. Do you have a favorite product that you’ve recently discovered that you just really love? It could be an app, it could be something physical, it could be a water bottle.
Nabeel S. Qureshi: I don’t have a good answer to that one. I guess I don’t buy enough stuff.
Lenny Rachitsky: Fully acceptable. There’s no wrong answers in the lightning round. Moving on, do you have a favorite life motto that you often find useful in work or in life that you come back to, that you share with friends or family?
Nabeel S. Qureshi: So there’s this architect called Christopher Alexander who wrote these beautiful books that are about beauty and kind of more than architecture. And he was a teacher at UC, Berkeley, and he got really frustrated with the students because he just felt like they were always turning in kind of average work. And so he would always tell them every week, imagine there’s a gothic cathedral in France called Charge. And he would say, you have to aim for Charge. You have to make something that is better than that. That should be your goal, not to just turn in something that’s what you feel is good enough. You actually have to try and be better than the very, very best that ever did it. And I find myself just repeating this a lot to myself. It’s just aim for that, really try and do that. Otherwise, it’s very easy to anchor on something right in the middle. And you do this unconsciously all the time.
Lenny Rachitsky: So is that the motto, just aim for Charge?
Nabeel S. Qureshi: Yeah, yeah, yeah.
Lenny Rachitsky: I love that. Most people have no idea what that would be, but with the context is quite powerful. Final question, what’s a classic novel that `you think would be most valuable for product builders?
Nabeel S. Qureshi: My favorite novel is Anna Karenina, and I would recommend that everyone read Erica.
Lenny Rachitsky: I’m reading that right now. I’ve never read it before.
Nabeel S. Qureshi: No way. And yeah, so it’s by Leo Tolstoy. It’s this epic 19th century Russian novel that follows a set of characters across society. And I think it’s just extraordinary because what’s amazing about him is he’s just able to imagine himself into the brain of anybody. And so even … he will briefly just go into the consciousness of, I don’t know, the servant who’s bringing the meal to the table or something like that. And he’ll just tell you a page of what they were thinking, and then he’ll just flip back into his main character’s head. And I think that is the most impressive demonstration of this kind of skill I’ve ever seen.
And I think, to connect it to your question, this is what you have to do if you’re going to be really good at product, is you have to really think yourself into the other person’s head, and you have to be really seeing it the way that they do. And it’s so hard, especially as a founder or product person, not to just get stuck on your own way of seeing the problem, right? You wrote up this doc, you made these marks. You’re like, this is going to be great. And then you take it to somebody, they don’t care that much. You really have to exercise your empathy and understand why they see it that way and what they actually care about.
Lenny Rachitsky: What a beautiful way to bring it all together. Let me also add, while I’m reading the book, something … a tip here is, people talk about having Chat GPT voice mode, just kind of sitting there next to you. I found that extremely helpful with this book where I just ask what the hell does this thing mean? There’s all these Russian dances and balls and etiquette. You just ask and you’re like, I’m reading Anna Karenina, what does this mean? And it just tells you.
Nabeel S. Qureshi: Yes.
Lenny Rachitsky: So there’s another cool tip for AI. Okay. With that, Nabeel, this was incredible. Two final questions in case people want to look you up. Where can they find you online and how can listeners be useful to you?
Nabeel S. Qureshi: Find me online, my website is nabeelqu.co and my X handle is Nabeel QU, I’m probably most active on that, but yeah, my website has all the links and a bunch of essays and interesting stuff. How can you help me? I would say send me an email. My email is on my website. Introduce yourself, say hi. I love meeting people. I don’t always have time for coffees nowadays or things like that, but I genuinely do get a lot of energy from just receiving emails from interesting people, so please do reach out.
Lenny Rachitsky: Awesome. Definitely check out Nabeel’s Principles. Is that the name of that post?
Nabeel S. Qureshi: Yeah.
Lenny Rachitsky: Great. Okay. That’s one to start with, and then also there’s the Palantir Post that we just talked through. Okay. Nabeel, thank you so much for being here.
Nabeel S. Qureshi: Thank you. Appreciate it, Lenny.
Lenny Rachitsky: Bye everyone. Thank you so much for listening. If you found this valuable, you can subscribe to the show on Apple Podcasts, Spotify, or your favorite podcast app. Also, please consider giving us a rating or leaving a review as that really helps other listeners find the podcast. You can find all past episodes or learn more about the show at lennyspodcast.com. See you in the next episode.
Glossary
| English | 中文 |
|---|---|
| AIP | AIP |
| Andrew Luo | Andrew Luo |
| Anthropic | Anthropic |
| AR-Labs | AR-Labs |
| Asana | Asana |
| Barry McCardel | Barry McCardel |
| BD (Business Development) | 业务拓展 |
| bull thesis | 看多逻辑 |
| Charge | 沙特尔(法国沙特尔大教堂,Cathédrale Notre-Dame de Chartres) |
| Christopher Alexander | Christopher Alexander |
| CMO (Chief Medical Officer) | CMO |
| COO (Chief Operating Officer) | COO |
| Databricks | Databricks |
| David Hsu | David Hsu |
| Decision to Leave | 《分手的决心》 |
| EOP | EOP(评估与选择流程) |
| forward deployed engineer | 前线部署工程师 |
| Foundry | Foundry |
| GoCardless | GoCardless |
| Goodhart’s law | 古德哈特定律 |
| Google Docs syndrome | Google Docs 综合征 |
| Gotham | Gotham |
| Habermas | 哈贝马斯 |
| Head of Product | 产品负责人 |
| Henriad | 亨利四部曲(Henriad) |
| Hex | Hex |
| High Output Management | 《High Output Management》 |
| Impro | 《Impro》 |
| Intercom | Intercom |
| IPO (Initial Public Offering) | IPO |
| Leela | Leela(开源神经网络象棋引擎) |
| Lenny Rachitsky | Lenny Rachitsky |
| local maximum | 局部最优解 |
| Looker | Looker |
| Max Levchin | Max Levchin |
| Mercatus Center | Mercatus Center |
| murder board | 质询板 |
| National Institute of Health (NIH) | 国立卫生研究院 |
| North Star | 北极星 |
| OneSchema | OneSchema |
| Ontology | 本体论 |
| Operation Warp Speed | ”曲速行动”(Operation Warp Speed) |
| outbound email campaign | 主动邮件推广活动 |
| PayPal | PayPal |
| PD (Product Development) | 产品开发 |
| Peter Thiel | 彼得·蒂尔 |
| PM (Product Manager) | 产品经理 |
| product leverage | 产品杠杆 |
| Ramp | Ramp |
| Retool | Retool |
| revenue per engineer | 人均创收 |
| SaaS (Software as a Service) | SaaS |
| SAP | SAP |
| short form video | 短视频 |
| Shyam Sankar | Shyam Sankar |
| Snowflake | Snowflake |
| solutions architect | 解决方案架构师 |
| SVP (Senior Vice President) | 高级副总裁 |
| Ted Mabrey | Ted Mabrey |
| ticket size | 客单价 |
| Tyler Cowen | Tyler Cowen |
| unicorn | 独角兽 |
| Vanta | Vanta |
| Watershed | Watershed |
| YC (Y Combinator) | YC |
| Zero to One | 《从 0 到 1》 |
Reformatted by reformat_english.py
在这篇深度对话中,前Palantir员工Nabeel S. Qureshi揭示了这家约两千亿美元市值公司背后的成长引擎。Palantir的人才培养成效堪称卓越:离职产品经理中三成选择创业,这一比例远超第二名的十八%;在产品负责人职位输出上同样稳居全球前三。这些数据背后,是一套独特的人才筛选与培养机制——候选人须先证明自己是能够深入客户一线的“前线部署工程师”,而非从传统PM路径晋升。
对话深入探讨了Palantir如何将务实精神与数据能力相结合,如何在看似矛盾的服务模式与软件规模化之间找到平衡。Nabeel还分享了这家公司独特的文化基因:对独立思考与广泛智识兴趣的重视,以及联合创始人如何以近乎理想主义的姿态吸引顶尖人才。对于关注组织如何系统性培养产品领导者与创业者的读者而言,这段对话提供了难得的内部视角与可迁移的洞见。
Palantir 如何打造终极创始人工厂 | Nabeel S. Qureshi(创始人、作家、前 Palantir 员工)
文字记录
Lenny Rachitsky: 离开 Palantir 的产品经理中有 30% 会创办公司。给我们描述一下那里的人是什么样的吧。
Nabeel S. Qureshi: 我觉得他们在筛选时特别看重几个特质。一是非常独立思考、不惧反驳的人。二是拥有更广泛智识兴趣的人。
Lenny Rachitsky: Palantir 的产品经理和传统产品经理有什么区别?
Nabeel S. Qureshi: 他们在这一点上极其严格:只有那些先以前线部署工程师(forward deployed engineer)身份证明了自己的人,才能成为产品经理。你基本上不可能通过其他途径成为产品经理。Palantir 有两类工程师。一类在核心产品上工作,是传统的软件工程师。另一类工程师会被派到现场——你大概周一到周四会实际去客户办公的大楼里,和他们一起工作,你会真的在那里有一张办公桌。这种工程师后来就被称为前线部署工程师。
Palantir 出产顶尖产品领导者
Lenny Rachitsky: 你有什么大多数人不认同的观点?
Nabeel S. Qureshi: 我认为这在科技界算是一个有些反共识的看法。
Lenny Rachitsky: 今天的嘉宾是 Nabeel Qureshi。Nabeel 是一位创始人、作家、研究者和工程师。他最近在 Mercatus Center 与 Tyler Cowen 一起担任访问学者,研究 AI 政策。他曾与国立卫生研究院(NIH)及主要临床中心合作,创建了全球最大的医学数据集。他也在英格兰银行工作过一段时间。他是 GoCardless 的创始成员兼商业发展副总裁,GoCardless 是欧洲最大的金融科技独角兽之一。
而与今天话题最相关的,是 Nabeel 在 Palantir 度过了将近八年时间,担任前线部署工程师,与美国联邦机构合作开展公共卫生项目,包括 COVID-19 应对期间的公共卫生服务,以及将 AI 应用于药物发现。不管你是 Palantir 的粉丝还是对他们所做的一切嗤之以鼻,Palantir 都是一家重要且快速增长的公司,正在源源不断地培养出令人惊叹的产品领导者——正如你将听到的,比世界上任何其他公司都多。因此,值得深入研究和理解这家公司。
我从未听过一场深入探讨他们如何运作、如何打造产品、如何招聘,以及如何从一家以服务为主的公司扩展为软件公司的对话。所以,我非常高兴能为你带来这次内部分享。在我们的对话中,我们将深入讨论:Palantir 到底是做什么的;为什么善于管理大量数据是他们成功的一个被低估的秘诀;审视他们创新的前线部署工程师这一独特角色,以及其他公司可以从中借鉴什么。此外,还有他们如何招聘、如何培养出优秀的产品领导者,以及大量关于与客户沟通、打造产品和创业的建议。
如果你喜欢这档播客,别忘了在你最喜欢的播客应用或 YouTube 上订阅关注。另外,如果你成为我新闻通讯的年度订阅者,你可以免费获得一整年多款优秀产品的使用权,包括 Superhuman、Notion、Linear、Perplexity、Granola 等等。详情请访问 lennysnewsletter.com,点击 Bundle。
话不多说,有请 Nabeel Qureshi。
[广告部分已跳过]
Nabeel 的 Palantir 经历与 PM 创始人数据
Lenny Rachitsky: Nabeel,非常感谢你来参加播客。
Nabeel S. Qureshi: 谢谢,Lenny,很高兴来到这里。
Lenny Rachitsky: 在今天的对话中,我想聚焦于你最近写的一篇文章,其中分享了你对在 Palantir 工作时光的反思。你在那里待了大概将近八年。我对 Palantir 特别感兴趣的原因是,我最近做了大量研究,考察哪些公司招聘了最好的产品经理、培养出了最好的产品经理,而 Palantir 在我的研究中一次又一次地出现。
我快速分享几个数据。我考察了哪些公司培养出了最多的创始人,尤其是从产品经理团队中走出的创始人,Palantir 以压倒性优势位居第一——离开 Palantir 的产品经理中有 30% 创办了公司。第二名是 Intercom,比例为 18%。我还考察了离开后立即在新岗位获得晋升的产品经理来自哪些公司,Palantir 在全球所有公司中排名第一。
我考察了哪些公司的产品经理加入其他创业公司后成为那里的第一位产品经理,Palantir 在全球排名第二。然后我考察了哪些公司的产品经理校友在日后成为产品负责人(Head of Product),Palantir 在全球排名第三。而且,这家公司的业务本身也做得非常好——目前市值大约 2000 亿美元。所以,从 Palantir 身上有很多值得学习的地方。
我想以一个 Palantir 每个员工大概都经常被问到、而我觉得很多人至今仍没能在心里给出一个清晰答案的问题开场:Palantir 到底是做什么的?
**Nabeel S. Qureshi:**这个问题很好。Lenny,你一上来就挑了个简单的。Palantir 呢,我的描述方式是——他们以非常务实的方式为客户达成成果。实现的方式通常是通过一个数据平台。所以,他们拥有我认为全世界最好的数据平台,我稍后可以展开讲讲这意味着什么。然后这个平台有几个不同的版本:一个针对情报和国防场景做了优化,叫 Gotham;另一个更针对商业场景,叫 Foundry。
这是对他们业务最经典的解释。他们卖的是一个数据平台。另外一点是,他们通常合作的都是非常大的客户——财富 50 强级别的,全球各国政府,诸如此类。这是简要的回答,但里面有很多值得展开的。
**Lenny Rachitsky:**太好了。我们后面会谈到很多这些内容,包括数据部分。我想先聊聊 Palantir 的人和公司文化。你分享了很多有趣的故事,讲在那儿上班甚至面试是什么体验。有一个你说到的故事——可能是因为某位联合创始人路过的时候在嚼冰块,说这对认知能力有好处。给我们描绘一下那里的人是什么样的,特别是 Palantir 早期,以及它的文化有多么独特。
Palantir 的人才筛选与文化特质
**Nabeel S. Qureshi:**是的,这确实是一家独一无二的公司。我觉得除非你是彼得·蒂尔那样的人,否则根本没法创办这样一家公司。具体来说,他们曾经在 Palo Alto 占据了夸张比例的办公空间。你走在 Palo Alto,到处都是 Palantir 的卫衣、Palantir 的办公楼,等等。
我觉得后来发生的事情是,他们融了大量资金,然后想出了各种非常有趣的方式,从 Stanford 等顶尖高校挖走顶尖人才,以及那些认识创始人的人——而这些创始人本身都是非常有趣的知识分子。我觉得他们特别着重筛选几个特质。
第一,非常独立思考的人——不怕反驳、敢于质疑一切框架、有自己的判断、立场鲜明。
第二,知识兴趣更广泛的人。Karp 刚出了一本新书,他在里面引用哈贝马斯和各种欧洲知识分子,这些你在一般科技公司 CEO 身上是看不到的。所以我觉得公司里有这样一条知识分子的基因。第三,就是竞争意识极强的人。公司有一种不惜一切代价要赢的心态。所以这些特质在当时形成了某种引力场。那个时期,大量极其有趣的人加入了这家公司。
他们筛选这些特质的方式也很有意思。在很长一段时间里,他们有一条规矩——现在我觉得大家都这么做了,但当时还比较少见——就是必须有一位创始人亲自面试你,你才能拿到 offer。创始人可能是 Alex Karp,可能是 Stephen Cohen,更早期可能像 Joe Lonsdale 那样的人,但一定是其中之一。
而且面试本身非常奇怪。跟 Stephen 面试,就是聊一个半小时的哲学,他基本上会凭空挑一个话题。完全没法准备,然后他会非常非常深入地试探你理解的极限。但这其实就是一场很有趣的对话,如果你过了感觉这关,你就进了。所以这是一种很强的筛选机制。
“坏信号”与使命吸引力
还有一点,我觉得可能是蒂尔提到的——他认为世界上最好的招聘者,或者说那些最能吸引人才的公司,他们都会发出一种独特的”坏信号”(bad signal),而且它必须让一些人反感。一个好的”坏信号”的关键恰恰在于此。我觉得在当下,OpenAI 和 Anthropic 都在吸纳一些你我都认识的最顶尖的人才。他们的一种方式——他们在这一点上是真诚的——就是确实吸引了那些对人工超级智能的潜力近乎弥赛亚式信仰的人,那些真正相信这是唯一重要的事情、将成为世界上最大事件的人。
我觉得 Palantir 当年的版本是,他们非常关注诸如捍卫西方这样的议题。有一个口号叫”Save the Shire”(拯救夏尔),对吧?所以他们早在所有人之前就在谈论军事、国防、情报以及这些事情的重要性。要知道,那还是在社交、移动、本地应用盛行的时代。社交媒体正在崛起。当时的热门公司是 Facebook、Pinterest 之类的。所以在那个时候,Palantir 做的事情非常奇怪。
所以我想,被这些吸引的人,必须看着其他选择说,“这些也不错,但我人生真正想做的是什么?“而另一边有个地方在说,“嘿,来和我们一起解决世界上最难、最棘手的问题。“我觉得正是在那个时期,这吸引了一些非常优秀的人。
**Lenny Rachitsky:**我们后面会谈到人们不太喜欢 Palantir 的原因以及他们所做的事情涉及到的道德问题。但当人们看到一家公司——我觉得你说的 OpenAI 就是个很好的例子——被他们的做法所反感时,你没意识到的是,这可能恰恰是有意为之的,因为它实际上吸引了他们真正想要的人。
这让我想到,我曾参与制定 Airbnb 的核心价值观。在那个过程中我们学到的一件事是,当你为公司定义价值观的时候,非常重要的一点是要明确这不适用于谁——正如你所说的那样。这感觉很反直觉,比如”哦,我们要包容,我们不想让人觉得他们不属于这里。“但核心理念是要清楚地界定谁会在这里蓬勃发展,谁与我们的使命一致。而我现在听到的是,Palantir 和这些公司把这一点做到了极致。
“质询板”机制
**Nabeel S. Qureshi:**完全同意。在我在 Palantir 的团队里,我们遵循一个流程——如果大家感兴趣我可以多讲讲——就是当你启动一个新项目时,你必须组织一个他们所谓的”质询板”(murder board)。我觉得这个词最初来自军队。大致做法是,你写一份两页纸的项目计划,然后邀请三四个对项目一无所知的聪明人,他们的工作就是把你的计划批得体无完肤。
所以你要写:这是项目的愿景,这些是目标,这些是未来三个月的策略。其中有一节是你这个项目遵循的原则。我记得我经常给新人一个建议——他们刚来的时候会写一些诸如”快速行动”之类的原则,我总会说,“大家都想快速行动。“这其实不是一个好的原则,因为没有人能合理地反对它。你需要的是那种很多人看了会说”你为什么要定这个原则?我觉得这不对”的东西。你需要的是人们可以反对的东西。
为什么”好胜”能筛选出对的人
**Lenny Rachitsky:**我想回到你一开始提到的,Palantir 在人身上寻找什么样的特质。你谈到了独立思考、兴趣广泛、有竞争意识。首先,我觉得很多人听到这些,尤其是最后一项,会想”我可不想去那儿工作。“但这套东西为什么有效?因为这并不是人们自然而然会想到的、打造最优秀最高效团队的方式。
**Nabeel S. Qureshi:**我觉得它就是吸引那些想赢的人。我认为这才是真正重要的。另一部分是,事实上——这一点在十年前更为明显——有大量人才稍微处于科技生态系统之外,但完全有可能在其中取得巨大成功。比如从军队或某个情报机构出来的人,他们可能正在某个地方读 MBA,准备转型进入企业界。通常情况下,他们会去一家经典的财富 500 强公司任职。而实际上,Palantir 成功吸纳了大量这类人才。在当时,这些人才是被严重低估的。
在海军陆战队或特种部队等地方最成功的人,往往相当聪明。他们往往在非常恶劣的环境中完成过极其困难的目标。事实证明,当你创办一家多少有些混乱的科技公司时,这实际上是一项非常有用的技能。当然,现在有更多公司在这么做了,比如 Scale AI 等等。但在当时,这是一个非常有差异化的人才池。
因此,我认为秉持那样的价值观——而不是当时更流行的那些,比如宣扬你多么包容,或者午餐供应的寿司如何如何——它吸引的完全是另一类人。我觉得这场博弈,第一层是使命对齐——你做的是一家国防公司,你想吸引的就是那种人。但我觉得还有第二层,就是哪些人才在当下可能被低估了,你如何真正把他们吸引过来?而这场博弈的答案一直在变。
Palantir 为何成为创始人训练营
**Lenny Rachitsky:**这确实开始解释为什么这么多 Palantir 的校友后来去创办公司,或在其他公司成为领导者。你们招进来的本身就是领导者。所以感觉很大程度上,就是因为你们招的人天生就是领导者。
**Nabeel S. Qureshi:**你说得对,我们可以再深入聊聊。但我觉得那家公司在非常具体的层面上确实是一个创始人训练营。我甚至认为它把很多本来不会成为创始人的人,变成了优秀的创始人,这归功于它的运作方式。所以我觉得那里既有选择效应,也有一些培养效应,但培养效应是这家公司独特的工作方式所带来的。
**Lenny Rachitsky:**这是不是就是前线部署工程师那套东西,还是别的什么?
**Nabeel S. Qureshi:**就是那个,是的。
**Lenny Rachitsky:**好,我们一会儿聊到那个。我很期待。好,太棒了。在此之前,还有最后一件事——我注意到你们在 Palantir 并没有真正的职级头衔。大家都是同一级别,所有人的头衔都是统一的。聊聊这个吧。你觉得这为什么重要?为什么有用?
没有头衔的公司
**Nabeel S. Qureshi:**我不完全确定是不是这个原因,但我确实知道蒂尔在《从 0 到 1》里写过这个。他的观点就是,一旦有了这些头衔,你就有了一个人们会去争夺的东西,然后就会产生非常低效的冲突。人们会为了钻制度的空子而优化。古德哈特定律(Goodhart’s law)无处不在。就是说,你设了一个指标,然后大家就都围着这个指标管理。
我不想拿某一家公司举例,但如果你看 Google,很多离开 Google 的人写了很有意思的帖子,他们把晋升机制列为心生不满的原因之一。获得晋升的方式是:与其改进一个现有产品,不如启动一个全新的产品,并且挂上你的名字。然后到了晋升季,你可以说”嘿,我做了这个新东西。“于是又多了一个 Google 产品,但对用户来说可能反而更困惑了。
所以我觉得他们想避免所有这类博弈行为。他们的做法就是说,“头衔不会成为人人争抢的那个名分图腾。相反,所有人都有同一个略显无意义的头衔,就是前线部署工程师。“唯一有正式头衔的人是 CEO,然后还有六位总监,仅此而已。现在我想稍微复杂一些了,有不同的团队,有些人有了头衔。但说实话,那时候几乎就是……
我们以前还拿这个开玩笑。就是有人离开公司后,你会看到他们更新 LinkedIn,写的是”哦对,我当时就是 XYZ 的高级副总裁。“然后你就想说,“不,你不是。你明明就是……”但话说回来我也完全理解,因为当你离开公司时,你必须让你的经历对外人来说是可理解的。所以你猜怎么着?像高级副总裁这种头衔确实是有用的。
所以,是的,我觉得他们想避免这种内部竞争。但这样做也有弊端。竞争也许不再围绕某个具体的头衔展开,但它变成了——有一个特定的项目主管或其他什么人,你想获得他的青睐。于是竞争变成了谁能进入某个人的核心圈子之类的。这种动态也是存在的。
不过我实际上很推崇没有头衔这种理念。我觉得它的确做到了一件事:假设你在负责一个非常重要的项目——这是完全可能的——它传达的信息是……这个角色始终是流动的。你在这个位置上是因为你很出色,所以这是一种精英制的安排。但如果你开始表现不好,实际上很容易就能把你换掉,因为不存在一个明确的”我是这个项目的总经理”的头衔。所以在公司里,你始终需要凭实力赢得自己的位置,你始终需要证明自己有资格做你正在做的事情。我认为这是一个好的副作用。
什么是前线部署工程师
**Lenny Rachitsky:**我们来聊聊前线部署工程师吧。什么是前线部署工程师?
**Nabeel S. Qureshi:**这个角色的起源,基本上你可以这样理解:Palantir 有两种类型的工程师。一种负责核心产品。他们不一定会离开帕洛阿尔托或纽约的办公室。他们就是非常传统的软件工程师,专注在核心产品上。
因为这家公司的运作方式——与这些大型机构有非常大规模的合作——所以需要另一种类型的工程师,被派到现场去。意思是,你可能在周一到周四期间,真正走进客户办公的那栋楼,和他们并肩工作。你会在那里有一张自己的办公桌。于是,这种工程师就被称为前线部署工程师。
前线部署工程师的两种类型
**Nabeel S. Qureshi:**所以在公司内部,这个职能叫作业务拓展,简称 BD。而 PD 是产品开发(Product Development),就是产品做出来的地方。在 BD 内部,你有前线部署工程师。其实有两种类型。一种偏技术,是更传统的软件工程师——你需要通过软件工程面试,证明你的技术实力,通常还会有计算机科学学位。但实际上还有一种前线部署工程师并不要求这些。你仍然会有技术面试,但重点不在于你是否懂某个具体的 C++ 算法,而更多是:你能不能对数据进行推理?我们最初并没有这个区分,但事实证明,有很多”技术邻近型”的人才——姑且这么称呼吧——在与这些大型组织或大型公司合作时,你确实需要这样的人在场。因为把你做的事情翻译成高管能听懂的语言,或者能够在会议室里驾驭社交动态,所有这些都是非常有价值的技能。所以招聘标准有所不同,更侧重于:你作为一个人的精明程度。但所有这些都统一使用前线部署工程师这个头衔,本质上就是与客户一起工作的工程师。
**Lenny Rachitsky:**好,为了让大家都特别清楚——很多人听说过 Palantir 有前线部署工程师这个概念,还有少数几家公司也做了类似的事情。这确实挺激进的。就像你描述的那样,你基本上在客户公司有一张办公桌。你之前跟空客合作过,我们待会再聊。先让它变得更具体一些:你在空客有一张办公桌、一台电脑、登录权限等等。你一周去他们办公室四天,跟他们的员工坐在一起,肩并肩地为他们打造产品——而大多数人的做法只是”跟客户聊聊天”,偶尔做一次访谈,开个 Zoom 会议,分享一些设计稿之类的。这个模式就像是那个的超级加强版。大概可以这样理解吗?
身临其境:前线部署工程师的工作日常
**Nabeel S. Qureshi:**是的,没错。我们确实大部分时间都在那里。这样的一个副作用是:第一,你真正做到了与客户的问题同呼吸、共命运,学会了说他们的语言。最终,他们把你当成自己人,你和客户之间建立了非常紧密的纽带。比如在空客,我会去飞机生产的工厂,或者坐在工作人员旁边,一起诊断飞机的问题之类的。后来我还与国立卫生研究院(NIH)合作,那是美国政府的一部分。我实际上在那里有一张门禁卡,跟公务员、生物学家、临床医生以及在那里工作的人一起共事。
所以,就像你说的,这确实是一种相当激进的模式。从商业角度来看,我觉得关键的一点是:Palantir 的平均交易规模非常大,动辄好几百万美元,这意味着你可以在客户获得的整体服务中把这种人力成本覆盖掉。而且定价是根据客户获得的价值来定的。
按价值定价的商业模式
举个简单的例子,如果你是空客,假设你的某架飞机出了问题需要修复,而修复这个问题对你来说值一亿美元——那定价就会以此为锚定。它不会按照”嘿,你在购买数据基础设施,跟 Snowflake、Databricks 或其他类似供应商差不多”的方式来定价。它的锚定方式是:这是你得到的成果。
而前线部署工程师的职责不仅仅是部署软件,也不仅仅是销售软件,而是真正解决问题。所以你必须在现场。你必须找到那些真正向 CEO 汇报这个具体问题的关键利益相关者,跟他们建立关系,赢得他们的信任。在某些情况下,你还需要创建新的软件,使其能够真正解决你面前的全新问题。
从问题到产品的循环
我有一些朋友跟我们的能源公司客户合作,他们不得不学习油井运作的方方面面。然后从中发现,对于这个使用场景来说,流式数据实际上非常有价值。于是,砰的一下,突然就诞生了一个能处理流式数据的产品,成为了核心平台的一部分。这就是那个模式:你了解问题,找出什么样的软件最能解决它,构建那个软件,用它来实现目标,最终把它整合到更广泛的产品套件中。
所以你就能理解为什么这是一个锻造创始人的好熔炉了。这其实也是我加入时的一个论点——我说,“假设我做了五次这样的训练”,实际上我做的比五次还多。但就算你只做了五次,在五个截然不同的环境中经历了这个循环,你实际上会变得非常擅长这套流程:走进大楼,赢得对方的信任,找到会成为你用户的人,跟他们谈论问题,确保你做的东西真的能解决这些问题,而不是白费功夫。
每周的快速迭代节奏
你会获得非常快的反馈和迭代循环。每周都有一个节奏:周一,你进去开会。周一晚上,你做出东西。周二,你拿给人看。周二你收到反馈。周二晚上你迭代。周三再拿给人看。周三晚上再迭代。每周你能完成四到五个这样的循环,而且你的推进速度快得惊人。六周之后,你突然发现:哇,这东西真的很有价值,有人愿意为此付两千万美元,砰。我觉得这就是为什么从同样的流程中走出了这么多创始人。
**Lenny Rachitsky:**我现在非常清楚为什么这么多创始人从 Palantir 出来了。好的,正如你所描述的,这里面一个重要的要素是:先为空客或某个政府机构这样的客户构建一个一次性的解决方案来解决真实问题,然后从那当中创造出 Palantir 可以卖给其他公司的产品。更妙的是——客户付钱让你为他们解决这个问题,而这就等于在资助 Palantir 后来可以卖给所有人的另一个产品。多么巧妙的生意。
从服务到平台
然而在 Palantir 早期,所有人都觉得它只是一个服务公司,或者只是给空客这样的公司做软件的顾问——他们不可能把这个做成一个适用于很多人的平台。显然,他们确实在做这件事,而且成功了。这就是圣杯——解决一个客户的问题,然后卖给其他所有人。基本上每个 SaaS 公司都想这么做。你觉得是什么让他们真的做到了这一点并且做得这么好?有哪些行之有效的原则?
**Nabeel S. Qureshi:**好问题,确实如此。我觉得从我加入一直到 IPO 之后的一段时间,总会有人跟我说:“嘿,这不就是一个咨询公司吗?不就是一家咨询公司伪装成产品公司吗?“但最终事实变得不可否认。其一,我总是觉得好笑——当有人问”Palantir 到底做什么?“的时候,你其实可以直接上 YouTube 搜索 Palantir demo,能找到大量展示这个软件长什么样的演示视频。虽然没多少人知道这个,但你现在就可以拿信用卡注册并开始使用它。
**Lenny Rachitsky:**我可以拥有一个 Palantir 账户?
**Nabeel S. Qureshi:**确实可以。
**Lenny Rachitsky:**我都不知道,太酷了。
**Nabeel S. Qureshi:**对,我想它现在叫 AIP 了。所以它其实并没有那么神秘,确实存在一个产品,如果你看利润率,它也能证明这一点——他们的利润率在 80% 以上,如果你真的是一家咨询公司,不太可能达到这个水平,更接近 20% 到 30%。那么你的问题是,他们到底是怎么做到的?我认为产品开发组织中确实有极其出色的人才,真正顶尖的、不可思议的人才。需要一些非常非常聪明的人,把我们当时用来为客户创造价值的那些内部工具拿过来,然后思考:这个的统一版本是什么?如果这做成产品会是什么样子?在我所看到的过程中,Foundry 就是这样诞生的。我想 Gotham 在更早的时候也经历了类似的过程。但基本上这个过程是这样的——你一开始进去的时候,基本上就是带着 Jupyter Notebook 和一些数据集成工具,但都非常原始,你必须用这种方式来创造价值。
但我们持续构建对前线部署工程师有用的工具,所以我们自己就是自己的第一批客户。到了某个时候,有人提出了一个想法:“等等,如果我们把内部工具给客户用呢?“我记得当时这是一个非常激进的想法。然后 Shyam Sankar,我记得他是 CTO,也许现在已经是总裁了,他直接下令:“好,每个客户部署都必须在三个月内让客户用上这个。“当时这非常痛苦,因为这些工具是为那些书呆子般的硅谷工程师构建的,所以并不特别好用,经常崩溃,你得去调试 Spark 报错之类的问题。但基本上这个过程让我们的产品思维变得更加严谨。
经过这样一个大约三四年的过程,Foundry 产品诞生了。之后又围绕性能和可靠性等方面进行了大量投入,那些都很痛苦。所以是的,我认为答案就是人才。另外还有一个认知,就是我们确实知道大多数人不了解的关于大型组织中数据如何运作的知识。这是另一件事——在与客户长期共处的过程中,我们发现了许多”秘密”。
最基本的一个就是,组织内部的数据集成(data integration)极其痛苦。除非你在大型组织中工作过,否则很难理解,但事实上即使到现在,你也很难获取到自己工作所需的很多内部数据。你会听到这样的故事:“我在计算我们这个季度的销售额,但不得不等了六个星期,等某个分析团队给我交付这个数据。“所以了解这类问题,并能够将我们的产品努力集中在这些问题上,意味着我们能够构建出可推广的解决方案。
Foundry 与 Gotham 的区别
**Lenny Rachitsky:**好的,这里面有很多内容。首先,你提到了 Gotham 和 Foundry。我知道我们会附上人们评测这些产品的视频链接,但用最简单的方式理解这两个产品各自做什么,你会怎么说?
**Nabeel S. Qureshi:**Gotham 针对军事和国防用途进行了优化,情报领域也是。我觉得它们有一些共同之处——两者都有一个,我几乎可以将其描述为金字塔结构:底层是数据接入(data ingestion),中间层是数据映射(data mapping),顶层是所有面向用户的部分,也就是所有的 UI 组件。如果你想一下 Foundry,有不同的工具可以让你接入数据,有不同的工具可以让你轻松构建数据管道和清洗数据——这是每个人都要做的——然后还有一堆工具让你在上面构建有吸引力的 UI 界面,做点选式分析,做 Notebook 风格的工作流,取决于你的技术水平。所以它作为一个平台,是一套共享统一数据底座但包含多个不同应用的工具组合。我想 Gotham 在某种程度上也是如此,但当你登录时,你会看到一个统一的界面。
那么两者的实际区别是什么?我认为在 Gotham 中,你会看到更多涉及地图的工作流。比如你在做军事行动的时候,很多时候你要看着一张地图,监控部队或坦克之类的移动。另一个重大区别是基于图的分析(graph-based analysis)的理念。Gotham 的一个用途是在恐怖分子网络中进行梳理,基本上就是找出坏人,所以能够执行基于图的查询很重要。比如,“Lenny 过去一周给每个人都打了哪些电话?“想象一下从那里展开的所有节点。然后,“好,这个看起来很有意思,我们放大看看。这个人位置在哪里?”
所以这是一种非常基于图的思维方式,也适用于欺诈检测等领域。Gotham 已经被部署用于反欺诈。但如果你看 Foundry,它实际上并没有那么强调这个组件,因为事实证明,假设你是一家 B2B SaaS 公司,你可能并不怎么需要基于图的分析,你做的事情更像是经典的 SQL 查询、表格之类的。所以 Foundry 在这方面更加传统。
**Lenny Rachitsky:**这个解释太棒了,我第一次开始理解这些产品到底做什么了。基本上就是吸收大量数据,清洗干净让你可以真正信任它,然后帮助你在各种场景下与数据交互——地图、图谱、表格。
**Nabeel S. Qureshi:**是的。
空客项目的故事
**Lenny Rachitsky:**太好了。你举了在空客做的那个项目的例子,你把它描述成基本上就是一个造飞机的桑拿房,是这样吗?
**Nabeel S. Qureshi:**是的。
**Lenny Rachitsky:**那么其中有多少成为了这个核心产品的一部分,又有多少仍然是一次性的东西?是不是有一些元素——某个很酷的创新——被放进了 Foundry?这个过程是怎么运作的?
**Nabeel S. Qureshi:**这其实是一个非常有意思的故事。我们最初进入空客时面临的问题是,他们有一款新飞机叫 A350,非常漂亮的飞机。顺便说一下,如果你坐纽约飞新加坡的航班,经常就是 A350,真的很不错。当时它还算是比较新的飞机,他们给我们的任务是:“好,我们需要快速提升这款飞机的产量”,比以往任何时候都快得多。数字是非常粗略的,但大致是:“好,这个月我们生产 4 架,下个月需要做到 8 架,再下个月 16 架,以此类推,你们要帮我们做到。“这又回到了我之前说的——任务不是”嘿,我们需要升级数据基础设施,觉得你们符合需求清单”,而更像是”请帮我们完成这个使命,这才是最重要的事情。“
为空客构建工厂管理工具
**Nabeel S. Qureshi:**于是我们进驻,梳理了这个问题。有很多不同的东西可以构建来帮助加速生产,但我们发现的一个基本问题是——不钻太多技术细节——工厂的运作方式是这样的:有一系列工位,你可以把飞机想象成在工位之间逐个移动,每个工位对它执行一组特定的工作。所以最初,它就是一个大的机身,机身停在那里,然后人们对着它执行一堆工单。他们需要零部件来完成这些工作。到了某个时刻,他们会说:“好了,这个可以转到 31 号工位了”,然后飞机被物理移动到下一个工位,31 号工位开始做它的事情。
所以为了让下一个工位能正常工作,他们需要知道:第一,上一个工位做了什么工作,还剩什么没做?第二——如果你想想这个问题——并非所有工作都能按时完成。所以会有工作结转到下一个团队,下一个团队就不得不……所以当我向你描述这个问题时,你可以开始想象:好吧,也许我需要某种甘特图来管理这些,我需要能够点进去看:“好,30 号工位做了什么,哪些工单还没完成?“然后再看:“针对这些工单,我需要哪些零件,它们在工厂的什么位置?“这个问题本身就已经非常非常难解决了。大量工作依赖于人们走到工厂车间里和其他人面对面沟通。对于来自科技行业的人来说——虽然科技公司的事可能不像造飞机那么复杂,造飞机确实是一个极其复杂的过程——但很容易看出,这个问题其实是可以用软件来改善的。
SAP 数据与本体论(Ontology)的诞生
所有那些数据都存储在 SAP 里。SAP 是成熟的软件,它擅长自己做的事,但不一定是最用户友好的,尤其是如果你不是它数据存储方式的专家的话。表名非常难以理解和阅读。所以我们发现的其中一个问题是:如果你能把那些表拉出来——那些表名基本上跟外星语言一样,表名就是类似 S3、F1_Z 这种东西——你得知道,好,这个表是存零件 ID 的之类的。
如果你能把这些表拉出来,用正确的方式做关联,然后把它们映射到人类能理解的概念上——比如零件、工单、飞机等等——基本上在它们之间建立一个层级或映射关系,那么用户就可以登录说:“好,79 号飞机在哪?好,在 31 号工位。好,这些是相关的工单等等。“所以你把它翻译成了人类可读的东西。
所以我们构建的东西——我有点随意地把它描述成 Asana——其实有点不同。但基本上它做的就是给你一个统一视图:好,这是工厂里正在发生的事情,这是这架飞机需要完成的工作。然后作为今天在 31 号工位上班的工人,我需要完成哪些工单,我需要的零件在哪里?那么这个直接成为 Foundry 的一部分了吗?不完全是,因为其他公司不会使用同一套概念,但”拿一堆表,然后映射到人类可理解的概念”这个整体思路是一个非常有价值的想法。
所以这实际上催生了 Foundry 中一个很大的组件,他们称之为 Ontology。如果你看过 Palantir 的演示文稿,可能听过这个词——他们总是在讲 Ontology。他们指的就是这个:一套作为人类能够理解的概念集合,你不需要自己去翻找和挖掘。你只需要说:“飞机现在在哪,下一步要去哪?“Ontology 成为了 Foundry 中极其重要的一块。它直接源自我们在那个工厂里构建应用时的经验教训。而且我认为它至今仍然是一个很大的差异化优势。我觉得没有太多其他公司在做这类东西。
**Lenny Rachitsky:**哇,我喜欢你现在谈到这个还这么兴奋。我能想象解决这么大的问题一定非常有成就感。我看到一个数据,好像是生产力提升了 4 倍。具体数字是多少?
**Nabeel S. Qureshi:**对,我不记得确切数字了,但我们确实在那一年至少把产量提升了 4 倍。当然,这是他们做到的,我们只是协助。但那位 CEO 说我们发挥了关键作用。
**Lenny Rachitsky:**而且你为此搬到了法国,对吧。这就是你作为前线部署工程师部署得有多前线的体现。你在法国住了多久?
**Nabeel S. Qureshi:**对,我在法国住了大约一年半。他们制造飞机的方式是在欧洲各地生产不同的部件。所以他们在西班牙制造尾翼,在英国和德国的部分地区制造机身等等。然后基本上把所有东西运到法国进行最终总装,你可以想象这是一个非常复杂的过程。所以我主要在法国,但有些周我得在这些国家之间飞来飞去,就为了搞清楚东西在哪里。
**Lenny Rachitsky:**你在文章里写到,前线部署工程师的生活真的很疯狂。有时候就是接到一个电话:“嘿,你明天要飞去某个国家,准备好。“这就是前线部署工程师的日常吗?
**Nabeel S. Qureshi:**是的。公司在差旅方面有一种非常——我会说——激进的态度。你入职的时候,基本上会被明确告知:“听着,你必须能接受出差,你能接受吗?“这种态度——我认为这也是一种非常创始人友好的思维方式——就是你必须愿意当天晚上就跳上飞机,如果那是对客户最好的选择,如果那是能让我们赢得这场战役的必要之举。所以有很多次就是这样:“我明天需要飞一个洲际航班去处理这个事情,因为这会有帮助。”
所以我的一个收获就是:当你和外部方合作时,亲自到场是极其有价值的。哪怕只是去那里待几天,花时间和他们在一起,也许一起吃顿晚餐。你能建立的信任远比通过 Zoom 去谈客户或远程推进项目要多得多。整个氛围完全不同。所以,跳上飞机曾经是我们工作中非常酷的一部分,持续了很长时间。当然这在 2020 年左右发生了变化,因为新冠疫情爆发了,公司也完成了 IPO,所以需要在这方面加强一些内部管控。但我想说在 2020 年之前,这是公司文化中非常重要的一部分。
**Lenny Rachitsky:**今天很高兴邀请到 Andrew Luo 加入我们。Andrew 是 OneSchema 的 CEO,OneSchema 是我们播客的长期赞助商。欢迎你,Andrew。
**Andrew Luo:**谢谢邀请我,Lenny。很高兴来到这里。
**Lenny Rachitsky:**OneSchema 最近有什么新动态?我知道你们和我最喜欢的一些公司合作,比如 Ramp、Vanta 和 Watershed。听说你们推出了一个新的数据导入产品,可以自动化团队在导入、映射和集成 CSV 及 Excel 文件上花费的大量手工工作?
前线部署工程师的适用边界
**Lenny Rachitsky:**有很多创始人在听这期节目,我有一个问题,他们大概也在想。其实这里有两个问题:一个是自己的前线部署运营要做到多深入。第二个是——我认识的一家公司确实在这么做——在单个公司的问题上投入多深,就是说”我们要彻底拿下这一个客户的问题”,同时赌这东西能抽象出来,做成一个大平台卖出去。那就从这里开始吧。你在创办公司,对于”我们先解决第一个客户的问题,赌这会成为面向很多其他公司的大机会”这条路要走多远,有什么洞见或建议?
**Nabeel S. Qureshi:**关于前线部署这块,我的朋友 Barry McCardel——数据分析公司 Hex 的 CEO——他其实写过一篇很好的文章,核心观点就是”你可能并不需要前线部署工程师”。非常具体。但我觉得关键在于,你必须愿意近乎浪费地去投入。你必须愿意投入大量资源去找到那个东西。而要做到这一点,你需要一定的客单价。所以每个客户的收入大概要在数十亿美元的规模。如果低于这个水平,你大概率看的就不是传统的前线部署工程师模式,而是有些不同的东西。
我认为很多从 Palantir 出来创业的人都带着一个论点:有很多 Palantir 不愿服务的客户,可能因为客单价太小。所以实际上你可以去为这些公司做一个类似 Palantir 的东西,但不是收 500 万美元,而是 25 万美元。在那种场景下,你可能仍然有前线部署工程师,但他们不会飞到法国每周五天待在工厂里。更可能是你安排一个人,同时照看五个不同的客户账号。大概是这样的比例才能让账算得过来。所以很多原则确实可以从那种经验中抽象出来,但这确实是一个非常特定的销售模式,依赖于一种特定的做生意的方式。
产品愿景与客户需求的抉择
关于你的另一个问题,我觉得这显然很难给出一个通用答案。我的核心观点是,你一定能分辨出什么时候你只是在做咨询,什么时候你更接近于在做一个产品。而人们常犯的错误,更多时候其实是过于固守自己的产品愿景。我看到的这种错误实际上比反过来的情况更多。如果你去找一个企业客户,假设你觉得你在做分析软件,结果发现他们其实没那么在乎内部分析,他们实际上有另一个巨大的棘手问题,而且还没有好的解决方案。很多人不愿意转向那个大问题,因为他们会想”我们是做分析软件的,也许这个客户适合我们的东西”,也许这个判断是对的。在某些场景下,这确实是对的,你应该去找一个你的产品更能引起共鸣的不同客户。
在另一些场景下,正确的决定反而是转型,把所有资源押在那个大问题上,然后再去找其他有同样需求的客户。这里没有硬性规则。我记得读到过一篇很有意思的文章,应该是 Retool 的 David Hsu 写的,他说的情况正是如此。他好像也在 Palantir 工作过一段时间。他说他们有 Retool 这个产品,但完全没有获得任何关注。然后他尝试了一次主动邮件推广活动,仅仅把邮件主题改成了”轻松构建内部工具”。突然之间,他们开始收到大量 CTO 的回复,说的都是”对,这确实是我的一个巨大痛点”。
但完全相同的解决方案,他们之前的定位好像是”超级增强版 Excel”之类的,根本没人买账。所以他们只是改变了定位方式,找到了另一批买家,就这样成功了。所以确实没有硬性规则,但我认为你脑海中始终需要有一个选择矩阵,并且非常审慎地想清楚你正在走哪条路以及为什么。
**Lenny Rachitsky:**我觉得你那点建议非常重要。按照你的经验,通常人们是过于偏向哪一边了?比如,客户现在让我做的事情,并不是我认为他们需要或者更广泛的客户群体需要的东西。你的意思是说,实际上客户更可能是对的,也许你应该更多地聚焦于此,而不是执着于你那个抽象的愿景和最初的想法?
**Nabeel S. Qureshi:**我觉得是这样的。人很难不被自己的经验和既有认知所锚定,这本身就是一个问题。我观察到真正优秀的创始人有一个特质,就是他们能够放下大量先入为主的假设,几乎把一个新机会当作一张完全的白纸来对待。然后想办法重新塑造产品来抓住那个机会,这就是你不会被困在局部最优解的方式。
**Lenny Rachitsky:**你的另一个建议也非常棒。人们听了可能会说,“我们请不起一个工程师坐在某个客户的办公室里给他们做东西。“但你的观点是,你可以安排一个人同时服务五个客户。他们不是全职驻场,而是在各个客户之间轮转。这几乎就像销售工程师,就像你说的那种闪亮的销售角色,他们帮助确保客户成功。我知道 Looker 是一个著名的例子,他们好像就叫前线部署工程师。你还知道其他公司有某种版本的前线部署工程师吗?
**Nabeel S. Qureshi:**有很多。我知道 AR-Labs 现在也在招聘前线部署工程师,他们在组建前线部署工程师团队,这件事是可行的。但我觉得会有一些关键差异。我不认为 Anthropic 会走进一个企业客户,从头开始为他们构建一个全新的解决方案。他们会做的是利用 Anthropic 的产品体系。现在很多公司都有了这个标签,但我觉得真正让人困惑的地方在于,它实际上意味着几种不同的东西。Ted Mabrey——我想他是 Palantir 的商业负责人——也写过一篇相关的文章,那篇也很好,值得一起看。
**Lenny Rachitsky:**假设有人说,“我想在自己的公司尝试这种模式,“你觉得有哪些要点是他们必须做对的?你刚才描述了人们对前线部署工程师定义的光谱,如果要尝试这种做法,你觉得最需要做到正确的是什么?
**Nabeel S. Qureshi:**让我们的模式成功运作的关键因素,第一,他们必须是真正的工程师,能够自己构建产品。这是一个非常大的区别。很多时候公司会说”这个人是前线部署工程师”,但实际上他们更像是一个解决方案架构师,或者他们并不一定在构建任何新东西,只是在那里倾听、想办法部署现有产品。他们没有被授权去做新产品。而 Palantir 说的真正激进的地方是,“不。你进去,如果你需要一个全新的产品来完成这件事,你可以直接去构建它。“我认为这是真正关键的区别。
另外一个我已经提到过的要素,就是亲自到场的价值,以及与客户建立深厚的个人关系。我确实认为优秀的创始人本来就会这样做。他们跟买家互发短信,下班后成为朋友,把他们当作自己想帮助的人来看待。我觉得这非常有激励作用,还能让你深入理解客户所在行业的业务,了解那些商业运作的动态。举个简单的例子,美国的医院。人们很难直觉地把医院看作一个商业机构。人们觉得这是一个接受医疗服务的地方,但如果你从 COO 或 CMO 的视角去看它,那画面会非常、非常不同。
一个非常简单的例子——抱歉,这个有点黑暗——但就像餐厅希望尽可能快地翻台以最大化当天的收益?医院其实对病人也想做同样的事。他们希望治疗你,然后让你出院,腾出床位接收下一个病人。这并不是很直观,除非你认真思考那家医院的收入是怎么运作的。但一旦你想明白了,你就会发现,“这会带来一系列相关的问题。“然后你就会进入非常有趣的方向……
**Lenny Rachitsky:**就是那些行业里的术语和段子,能让你在理解和运作中走得很远。
**Nabeel S. Qureshi:**是的。
前线部署工程师的关键要素
**Lenny Rachitsky:**好,所以基本上你要做对的事情就是:确保是亲临现场,确保这个人是技术人员,确保他们对业务和客户面临的问题有深入理解。技术这个点在如今 AI 工具的加持下很有意思,从某种意义上说它让每个人都变得具备技术能力了。你可以说这会变得越来越普遍,人们可以直接打开 Cursor、Windsurf,然后就开始添加功能。
**Nabeel S. Qureshi:**我觉得你刚才触及的这个论点非常有趣,我预期会看到更多利用这个洞察的创业公司出现。
**Lenny Rachitsky:**本质上就是让前线部署工程师的成本更低了。
**Nabeel S. Qureshi:**没错。
Palantir 前线部署工程师的演变
**Lenny Rachitsky:**那 Palantir 现在前线部署工程师的状况如何?过去几年有多大变化?如果现在加入,这件事还能做吗?
**Nabeel S. Qureshi:**当然可以。我应该先强调,第一,我在 2023 年就离开了公司,所以这只是我的个人观点,我不代表公司发言。我觉得可以这样想:公司用来衡量自身成功的一个指标基本上就是人均创收(revenue per engineer),所以你的”产品杠杆”越高,这个数字就越大。如果你不得不为每一个边际问题投入大量人力,那你做得就不太好,因为你基本上每次都在构建一个新东西,实际上你就是一家咨询公司。但另一方面,如果你每遇到一个新客户,产品恰好就能满足他们的需求,那就很好。所以这个产品杠杆的指标实际上是非常独特的,在我整个任职期间一直是公司的北极星指标。
如果你按这个逻辑推理,这意味着在公司早期阶段,你可能有一个客户,然后有五到十个工程师在那个客户那里工作。随着时间推移,你希望这个比例发生变化。你希望因为产品足够强大,也许 AI 编码能力也有了很大提升,每个客户只需要两个人,然后也许你最终可以做到一个人同时照看多个客户。我觉得工作方式的变化就在于此——现在更多是你有多个客户,也许你在每个客户身上花的深度时间少了,但你跨多个客户解决的问题更清晰了,你也拥有了更明确的产品方案。
所以我认为这确实是一个变化,但公司仍然是一个非常有趣、充满活力的地方。在某种意义上,这个故事才刚刚开始,因为你可以这样看这家公司——他们花了二十年基本上在为世界上每一个重要机构构建一个终极数据基础设施。而现在 AI 模型出来之后,最有价值的东西就是非公开的专有数据。突然之间你拥有了这些数据的访问权,你就处在一个非常有利的地位,可以帮助客户以让他们成功的方式部署 AI,解决真正的业务问题。这本质上就是这家公司的看多逻辑,也是为什么它可能会再涨 100 倍。所以现在仍然是一个非常有趣的加入时机,但我确实认为人员与客户的比例关系,比如说,现在是一个很大的不同。
数据的力量
**Lenny Rachitsky:**这不是投资建议,但它确实可能涨 100 倍。我完全理解为什么这可能发生。我们来谈谈数据这个话题,你说过这是 Palantir 成功的秘诀之一,这个关于摄取数据、清洗数据、能够分析和利用数据的早期洞察。真是绝妙的营销,他们想明白了为什么这件事如此有价值、如此困难,以及他们是如何做到的。
**Nabeel S. Qureshi:**我觉得这件事只要你走进一家公司待上几天就会非常明显。你想,假设你的工作是提升销售额,那你首先要做的就是搞清楚目前的情况。好,那我去查一下销售数据库。等等,销售数据库在哪?我没有访问权限。好,我需要提交一个访问申请工单,然后等上一个星期。所以我们每到一处,这都是最大的痛点——光是获取数据访问权限就要等六到八周。而当你终于拿到权限的时候,数据也不是那种可以轻松查询的格式,你真的需要非常专业才能从中提取出正确的指标,诸如此类。
结果就是,好比一座冰山,真正的分析其实只是露出水面的冰山一角,大概就是最后那 5% 到 10%,而前面 95% 的工作是:获取数据访问权限、清洗数据、关联数据、标准化、把所有数据统一成相同的格式。一旦我们发现这一点,就会意识到这里面其实有很大的产品空间,可以让这个过程变得更容易。人们通常不认为 Palantir 是一个会产出创新产品和用户体验想法的地方,但我实际上认为在过去的 20 年里,它在这方面是最有创造力的公司之一,只是大部分成果并没有对外公开,所以大家不知道。但如果你看看他们为了让上述过程变得更简单而开发的产品原语,它们其实非常有价值、非常有趣,甚至可能成为独立公司的基础。
所以,一旦有了软件解决方案,这个过程的每一个步骤都变得容易得多。拿数据摄取来说,Foundry 内置了一个通用的数据适配器,几乎什么都能读取——JDBC、S3 bucket,随你怎么弄。它可以让你查看数据,预览前 20 行,等你准备好了还可以设置一个定时任务,按一定频率自动拉取数据。光是这一个过程,以前对工程师来说就要花很长时间,尤其是在 Vibe coding 出现之前,管理那些 cron job、在客户租户里的某个 analytics VM 上做这些事情,都是非常痛苦的。
你把这个环节产品化之后,接下来就是——数据有了,但怎么把它们关联起来?如果你不是技术人员呢?有没有办法让非技术用户也能做表关联,看到结果?所以就有了一系列非常有趣的商业问题。因为获取数据本身就很困难,人们之前并没有真正解决过,所以有很大的空白地带可以做产品创新。所以我敢说 Foundry 现在绝对是世界上最好的数据平台,就是因为它内部有这么多不同的应用,分别解决这些离散的环节。而这都来自于多年的痛苦经验——看着人们清洗数据、关联数据、搞清楚某个表名到底是什么意思,日复一日。
数据守门人
**Lenny Rachitsky:**你在文章中分享了一个很形象的故事——有些人的工作就是给数据当守门人,他们的职责就是控制你能否访问组织内部那些非常有价值的数据,获取数据真的很难。很多工作就是在打破那些政治壁垒——“我们需要这些数据,这是为了公司好”,这中间要花很大力气。关于这点你还有什么想补充的吗?
**Nabeel S. Qureshi:**是的,确实如此。这真的是一个非常大的痛点,而且也有其合理的原因。并不是说这些人有什么恶意。如果你是 IT 人员或者信息安全部门的人,你的目标是防止数据泄露、确保敏感信息不会扩散得太广。那最简单的方法是什么?就是把数据锁起来,基本上就是当数据的守门人。我觉得更有意思的地方在于,有些人的技能和岗位价值恰恰依赖于他们作为守门人的角色。我的意思是,假设我是唯一一个理解销售计算管线运作方式的人,相关的 SQL 都是我写的。业务专家们的所有请求都到我这里,我有一个很长的排队,要好几个星期才能处理完。我的工作很稳定,很有保障,大家都依赖我。
然后这家公司来了,说:“嘿,我们想把销售数据开放给所有人,让它变成点点鼠标就能用的东西。“你马上就会说:“等等,那我怎么办?“所以我觉得这里面的阻力非常大。我常说,人们问谁是竞争对手?我觉得不一定是你会想到的那些。Palantir 最大的竞争对手是客户自建方案——最大的分歧就是某个 CIO 说:“我要自己建数据基础设施,我自己掌控,建在某个超大规模云平台上,我们自己做所有的分析。“而我们的做法对此是非常颠覆性的,我们说的是:“不,你所有的数据都会被摄取到这一个平台里,你公司里的每个人都要用它。“代价是,每个人做事情都会变得非常非常简单。但你可以想象,有些人对这个模式并不买账。
招聘心得
**Lenny Rachitsky:**听你这么说我感觉 Glean 才是 Palantir 最大的竞争对手,你知道这家公司吗?
**Nabeel S. Qureshi:**我知道,Glean 从外面看确实很厉害。不过差别还是很大的,我完全理解你为什么这么说,但是——
**Lenny Rachitsky:**显然是不同的使用场景,但它成功的原因感觉就是它搞定了大量这种数据摄取、权限管理、搜索之类的东西。我之前从没从这个角度想过。
**Nabeel S. Qureshi:**对。
**Lenny Rachitsky:**有意思。好,我想聊聊招聘的话题,你之前谈过一些。你现在又在创业了,从 Palantir 的经历中学到了哪些关于招聘的关键教训?我不知道你现在是否已经在招人了,也许快开始了。
**Nabeel S. Qureshi:**是的,我们目前有六个人,团队还很小。说到招聘,挺有意思的——网上有那么多招聘建议,你看的时候会想”对,这些都很显而易见”。但当你亲身经历的时候,你会突然明白”啊,原来大家为什么都这么说”。举几个简单的例子。我觉得最难找到的是那种对做的事情真的非常热爱、愿意多付出那额外 20% 的人。我觉得当你向外招聘的时候——尤其不是要刻意针对谁,但我想如果你招的是那种[听不清]背景的人,他们想要的是一年 40 万美金的工作,想工作固定的时长,想写完代码就回家,基本上就是你在大型公司里习惯的那种模式,即便你本来并没有这个意图。
招聘中驱动力比技能更关键
**Nabeel S. Qureshi:**所以如果你从那样的人群中招人来做一家很小的创业公司,会非常困难,因为创业公司的成功很大程度上取决于每一个人都能说”不,我真的要——如果今晚需要把这个东西做好,我就工作到今晚,我不会只是走个过场打勾完成,我要去看这个企业真正想要达成的结果是什么”。我说的这些感觉都挺显而易见的,但当你真正感受到那种差别——一个只是打勾走流程的人,和一个在这方面简直像猛兽一样、会真正去追击并完成最终目标的人——这个差别非常、非常大,而且对你前 20 个人来说极其重要。而且找到这些人没有什么科学方法。不是说你只要在招聘启事里写上”关心结果”,然后忽然就会有一大堆这样的人来投简历。
所以问题是,好吧,那你怎么筛选,怎么找到这类人?这就变得很有意思了。我认为这就是使命认同感发挥作用的地方,你确实需要找到那些对你的事业有额外的、也许是私人的理由去比普通人更在乎它的人。比如 Palantir,他们确实招了很多退伍军人,或者那些可能比一般科技公司员工更爱国、更亲美的人,这些人对 Palantir 有额外的理由,有额外的动力去多付出那一点努力。我现在做的事情更偏向医疗健康领域,所以我觉得那些自己有过与这个系统打交道经历的人——也许有亲人经历过癌症之类的艰难遭遇——他们就是会多那么一分动力,真正在乎你试图做的事情,然后多付出那一点努力。所以我认为在早期要大力筛选使命匹配度——你过去有多在乎某些事情,举个例子,你问这样的问题:你曾经为了完成一件事拼到什么程度,为什么?这个问题确实能区分很多人,很多人其实答不上来。所以我觉得这是一个很大的经验教训——与其说是测试正确的技能,不如说技能当然重要,但更重要的是找到谁有那额外的 20%。
**Lenny Rachitsky:**这真的很有意思。你分享的所有内容本质上都围绕着动机、驱动力、热情,以及对专注投入的承诺,而”他们很聪明、技能很强”反而像是次要的考虑。感觉那只是基本门槛,真正决定性的是你说的这些东西。
**Nabeel S. Qureshi:**是的,我完全同意。而且我觉得每家公司的侧重点不同。如果你在 B2B SaaS 这样的领域,可能比较难讲出”这件事的使命如此重大”这类故事,那就有其他方式来捕捉这些品质。比如我认识很多人——虽然这个做法现在有点被用烂了——但对于销售团队,他们会明确去找那些曾经是职业运动员或大学时打过体育比赛的人。这测试的是什么?是你要非常有纪律,非常以目标和数字为导向,而且愿意非常非常努力地工作。所以有各种各样迂回的方式来捕捉这些品质,我觉得作为创始人你必须对此有意识地去做。拿我自己来说,我是个跑者,所以我其实很喜欢认识同样跑步的人,我开玩笑说”也许我会从跑团里招人什么的”。但这跟我下棋也一样,我经常下国际象棋,喜欢认识棋手。我不一定说那就是适合我的正确招聘方式,但我觉得拥有这样一些表面上看不太相关、但实际上能给你关于这个人性格的好的信号的特征,这些真的很重要。最后再说一件事,作为一个有趣的例子——Max Levchin 讲过一个故事,说 PayPal 早期有个人来面试,技能面试全部通过了,到最后一轮的时候,他说了句自己喜欢打篮球之类的,他们立马就拒了。那儿的氛围就是如果你不是一个超级 Linux 书呆子、硬核计算机人,那我们就不要你,哪怕你其实通过了所有测试,就因为你喜欢打篮球。至于这个决定是对是错,不好说,但这就是我说的那种例子。
**Lenny Rachitsky:**我觉得这个很好地呼应了前面说的。听到这个的人可能会想”搞什么?他们怎么敢这样?“但这恰恰是你在我们对话一开始说的——打造一家跨世代企业的方法,就是要非常清楚这不是为谁准备的,这没问题,这是你的公司,不是每个人都得来这儿工作。这其实也在帮他们节省时间,因为他们可能会意识到”这不是我想要的,这未必是我想共处的人”。所以我觉得看到这一面很重要——这是你的事业,说清楚谁是公司的好人选、谁不是,这一点很重要。
Palantir 的产品经理
**Lenny Rachitsky:**说到这个,我们来聊聊产品管理的话题。我知道 Palantir 的产品经理不是传统的产品经理。我想 Palantir 里应该有人顶着产品经理的头衔吧?如果是的话,据你了解,Palantir 的产品经理和比如说 FANG 公司的传统产品经理有什么区别?
**Nabeel S. Qureshi:**据我所知,Palantir 曾经很长一段时间是相当反产品经理的,后来我们确实需要了,因为我们开始在产品测试上变得更认真。
**Lenny Rachitsky:**经典故事,经典故事。
**Nabeel S. Qureshi:**经典故事。
**Lenny Rachitsky:**很多公司都这样。
**Nabeel S. Qureshi:**我注意到的一个重大区别是,他们非常谨慎地只让那些先在前线部署工程师岗位上证明过自己的人成为产品经理。你基本上不可能通过其他途径成为产品经理。举个例子,我之前提到过的我们为那个飞机工厂做的东西,管理那个部署的人后来成了本体论的产品经理,就是因为她已经在实地证明了自己的方法。原因很简单:这样的人理解客户怎么运作,有那种客户共情能力;而且是那种有驱动力把事情做成的人,因为业务拓展本身就是筛选这种品质的。我认为他们非常、非常抗拒的传统产品经理的失败模式,是一种”Google Docs 综合征”——好吧,我要写我的产品需求文档,我要用一种非常理性的方式来管理它。公司在这方面非常严格。
产品经理的培养路径
Nabeel S. Qureshi: 所以基本上产品经理几乎都是从内部晋升的,而且都来自业务拓展部门。我不记得有任何一例是我们从 Google 这样的地方招来一个产品经理——Google 培养出了很多优秀的产品经理——然后成功融入 Palantir 的。这就是一种非常不同的氛围。所以我觉得这是一方面。另外一点可能更偏经典的产品经理特质:你自己要么得是工程师,要么得非常擅长与工程师协作。我见过的最成功的那些产品经理,基本上都跟自己的工程团队是最好的朋友。团队通常就是一个所谓的”组产品经理”加上一群非常、非常优秀的工程师。成败的关键说到底就是:工程师信任你吗?我之前提到过 Palantir 的人性格上几乎都有点不太随和,所以如果你不能很快赢得工程团队的信任,你是待不长的。
Lenny Rachitsky: 我觉得我们已经解开了 Palantir 的产品经理为什么这么成功这个谜题了。首先,招聘门槛本质上就是在招领导者——在各个方面都是。这就好比一个创始人锻造营,他们在真实公司里解决真实问题,打造真正赚钱的产品,然后这些人就成了 Palantir 的产品经理,之后离开去创业——这就是为什么 30% 的人最终创办了自己的公司。说实话我很惊讶这个数字不是更高。要不然就是去其他公司做首任产品经理或者产品负责人。
Nabeel S. Qureshi: 对,确实很疯狂。我在 Palantir 加入的是一个相当小的团队,大概 20 到 25 人,而那个 25 人的小组里至少有六个人现在已经是独角兽或准独角兽公司的创始人了,这个比例真的很惊人。还有更多人最近在更早期阶段创办了公司,所以确实有这些一个又一个闪光的群体,看起来非常有意思。我觉得推动这种现象的另一个因素是:Palantir 是一家非常有趣的公司,所以留存率实际上非常高。人们在那里的任职时间通常比硅谷的平均任职时间长得多。所以当你决定离开的时候,通常是因为有某个非常具体的东西在拉着你,你想进入游戏的下一关。因此一个人离开后又去加入一家更传统的科技公司,这种情况非常少见。就好像——你要么去当创始人,否则你为什么要走呢?这里有那么多有趣的东西可以做。我知道这听起来有点像邪教,但那里的人确实都是这么想的。
Lenny Rachitsky: 我完全能理解。很多离开 Airbnb 的人再也没有找到更有意义的事情做,这确实很难,尤其是如果你是早期加入的话。还有一个数据我没提到,但我觉得非常有意思:如果你看看 YC 创始人的来源,我想你在文章里也分享过——前 Palantir 员工中成为 YC 创始人的人数,比前 Google 员工还多,尽管 Google 的员工基数大约是 Palantir 的 50 倍。
Nabeel S. Qureshi: 是的,没错。
Palantir 的道德争议
Lenny Rachitsky: 我们来谈谈 Palantir 的道德问题吧。很多人看到这期节目的标题、听到这些内容,可能不会高兴看到 Palantir 被这样展示和推广。很多人不认同 Palantir 做的事情——在某些方面它制造了能杀死人的产品,它与他们不认可的政府合作。我知道你在决定加入 Palantir 时写过一篇非常有洞察力的文章,谈到了你怎么看待这个问题、以及你观察到别人是怎么应对这个问题的。你能谈谈你最终形成的思考框架以及你自己的看法吗?
Nabeel S. Qureshi: 好,这是一个非常有意思的话题,确实有很多微妙的层面。我在那篇文章里想说的有这么几点。第一是,Palantir 确实做了很多有正面价值的事情。我参与了美国的新冠疫情应对工作,我有朋友参与了”曲速行动”(Operation Warp Speed),这些我认为都拯救了很多生命。我在国立卫生研究院工作期间主要专注于癌症研究。对我来说,这些事情显然是好的,而且你在其他地方做不到这些,仅凭这一点就有理由留下来。那篇文章中我提出的问题是:好吧,这家公司确实会有一些让人反对的地方。在 2016 到 2020 年那段时期,这已经变成了很常见的事——你去纽约上班,外面就有人在你的办公室门口抗议,做各种各样的事情。所以就产生了这样一个疑问:这样做对吗?我想表达的观点是,“不参与”很少是正确的答案。这一点现在可能更被广泛认可了,但在当时,风气确实走得太远了一些。
比如那个著名的例子:Google 仅仅因为一些人觉得与五角大楼合作本身就是不道德的,就退出了一个五角大楼的 AI 项目。我觉得这比大多数美国人的立场偏左了不少。大多数美国人会说,在合理范围内从事国防相关工作是可以的,前提是你在做的大部分是好事。所以在某个时刻,几乎出现了一种套利空间——等一下,从事国防工作并不是天生就邪恶的,它其实是一件相当有意思的事情。然后还有这个问题:你宁愿待在那个房间里推动事情变得更好,还是选择不参与?我在这里能分享多少有些受限,但举个简单的例子:即便是一个很多人可能不太舒服的工作流程——假设你在对某个目标进行打击定位——如果你把现在的做法和 2010 年的做法相比,现在会精准得多、准确得多,所以你实际上改进了那个流程,降低了出错的风险。也许你应该为此感到欣慰,对吧?当然,这是很多人不愿意接受的代价。
我自己没有在公司国防那边工作过,但我认为你必须能够接受这些灰色地带,并且真正主动地去思考自己在做什么。这并不意味着在国防公司工作永远是正确的事。也许我们会进入一个非常黑暗的未来,在某些方面我们开始变成坏人,那这时候在国防公司工作大概就不是什么好主意了。所以这是一个不断变化的局面,但我当时有一种很强烈的感觉:科技行业的很多人根本不愿意去思考这些问题。
你看现在有些工程师在做的事情是优化短视频以提高用户参与度,你差不多想对他们说:“嘿,你们有没有想过这对小孩子的大脑正在产生什么影响?“或者”你有没有见过一个 11 岁的孩子连续刷五个小时的短视频,你觉得这是一件好事吗?“我觉得人们不太愿意深入思考这些事情。我不是说我知道答案,但长期以来,科技行业几乎有一种拒绝从政治视角审视自身行为的风气。就是那种:“嘿,让我们玩自己的玩具,让我们待在自己的小公园里,别来打扰我们,我们就是要做酷东西然后发布出去。”
(01:14:08): 到了 2025 年,世界状态已经非常、非常不同了,科技现在已经深度参与政治,政治基本上也闯进了科技圈。有个很经典的画面,马克·扎克伯格坐在国会听证席上,看起来脸色苍白,仿佛在说”为什么又把我拖到这儿来了?“但我认为科技行业经历了一个这样的过程:哦,我们突然变得重要了;哦,我们真的、真的非常重要了;哦,我们最好别再玩这场政治游戏了。所以我觉得我现在说的这些,比起十年前已经得到了更多共识,但在当时,大家的感觉就是:“看,我们做的事本身就是政治性的,所以你最好正视这一点。”
Lenny Rachitsky: 我觉得这件事对很多人变得非常真实,是在俄乌战争的时候。政府用完了某些载具和弹药,生产跟不上,然后大家才意识到:“哦,谢天谢地还有 Anduril 这样的公司,以及其他真正领先、让我们保持领先地位的科技公司。“我认为美国在太空竞赛中领先中国,唯一的原因就是……
Lenny Rachitsky: ……保持领先地位。我觉得美国在太空竞赛中领先中国的唯一原因,就是 SpaceX 这一家公司长期深耕于此。所以我认为很多人已经开始意识到:好吧,也许我们确实需要这些东西。
Nabeel S. Qureshi: 没错。我也会提出同样的论点。很多人会问:“你怎么能在国防领域工作还感到心安?“但问题是,如果中国入侵台湾,你也不会好过的,对吧?实际上你肯定不会接受那个结果。所以我们确实生活在这样的世界里,确实需要建立威慑力量,而且这些力量必须足够强大。对我来说,这不是一个很难回答的问题。当然,聚焦到具体的事情上,确实可能非常困难,过去几年也出现过不少这样的案例。但归根结底,不参与不是答案。
Lenny Rachitsky: 是啊。而且再说一遍,这也不适合所有人。我觉得这其实是贯穿这次对话的一个重要主题:有时候要打造一家真正成功的跨代公司,你必须让一些人感到不适,因为往往正是这种特质才能吸引最优秀的人才。
(01:16:03): 好的,就再问几个问题。稍微退后一步来说。你现在又在创业了。你从 Palantir 的经历中带来了哪些核心建议,会影响你构建这家新公司的方式?我们聊了很多东西。有没有什么是你觉得”我一定会用这种方法,因为它在 Palantir 效果非常好”的?
Nabeel S. Qureshi: 一件事可能就是非常快的迭代周期。多下赌注,然后非常严格地快速走过这个循环。我有一些原则,其中有一条基本上就是:EOP 的成功率与你下的赌注数量成正比,这是下注数量和每个赌注成功概率的函数,对吧?所以一个几乎能保证你有所收获的简单方法,就是大量下注,然后快速轮转。当然这很困难,经常面临的问题是:这个赌注是不是真的失败了,还是我们放弃得太早了?但这是我的原则之一,就是尽早测试这个东西。就像经典的说”喂食、见真”的方法——当你把东西拿给客户看的时候,让他们付很多钱,然后发现下一个问题。不要等三周,而每个创始团队通常都会等三周,因为你们没有那么多时间。
(01:17:29): 我认为确实很重要的一点是,拥有一种非常紧密、独特的内部文化,在团队内部建立一种强烈的信任感。就像你提到 Airbnb 的那样,人们在 Palantir 也有这种感受,就是一种”你在那里工作过,那你一定很优秀。我信任你,所有这些。“我觉得创建这种氛围非常重要,而且你心里清楚那种感觉是什么。比如有人问我,应该去 X 公司工作,还是直接去创业?我不知道对每个人来说答案是什么,但我会说,在那样的公司工作有一个好处:你现在有了所有的内部基准,知道”好吧,这应该是这种感觉,如果不像是这种感觉,我们就走偏了。“我无法想象没有这些基准,一切都得自己摸索。
(01:18:21): 是的,我觉得同样重要的是,要建立独特、内部驱动、强大的团队文化。然后第三个,对我来说,就是去接触现实世界中那些非常复杂的领域。所以我离开时开玩笑说,我很高兴终于可以做纯软件了,我很高兴,我不知道,建一个 ID 系统之类的,甚至连客服邮箱都不用设。但结果你看,我的比较优势在很多方面其实在于我建立的这些人脉网络,以及我在接触现实世界复杂领域方面的经验。而这些领域确实非常需要技术,对吧?
(01:19:00): 我有时候会有一个令人不安的想法:也许两年后我们就能实现 AGI 了,但医疗行业仍然一团糟,在纽约还是租不起房,还是建不起房子,所有这些问题可能依然存在。这很可能会成为现实。所以我认为去接触现实世界中那些复杂的领域也很重要,尽管它们真的、真的非常具有挑战性。我认为关于 LLM 非常好的一点是,现在有太多工作流程对你作为科技创始人来说是可触及的,而且人们比以前更愿意与科技公司合作了。2015 年向这些经济部门销售产品,难上加难。而现在 ChatGPT 之后,人们愿意给小型创业公司机会,这是以前不会给的。正如你之前提到的,做前线部署工程师这类事情的成本可能已经下降了至少五到十倍。所以现在有很多新的可能性,我很兴奋能去接触最好的那些。
Lenny Rachitsky: 哇,这可是很有价值的信息——你发现了一些非常大的组织更愿意与创业公司合作。传统上投资者不愿意投资那些去拿下医疗公司和政府这类客户的公司。所以听到这个真的很有趣。
(01:20:17): 我来复述一下你刚才分享的建议,其中其实有一个次级建议,我觉得更有意思。第一个你要带走的是快速迭代,但我很喜欢你的建议:尽早索取很多钱,看看这是否真的是一个人们愿意付大价钱的想法。如果不是,就换一个。
(01:20:35): 另一个是建立非常独特的文化,但你分享的那个我更喜欢的部分是:知道高标准是什么样的,知道优秀的 A+ 级人才是什么样的,而你需要去 Palantir 这样的公司才能真正见识到。所以那里的建议我觉得其实是:先去一家出色的公司、与最优秀的人才一起工作,理解好的标准应该是什么样的,同时你还能建立这些人脉网络。所以我觉得这真的很有意思。
Nabeel S. Qureshi: 还有一条建议你也可以带走——去解决真正困难、复杂的问题,因为最大的机会就在那里,而且听起来这是实际从事这项工作的最简单时机。真的很神奇。
Lenny Rachitsky: 好的。现在带大家进入播客的一个常设环节——AI 角落。我们在这个环节会分享一些……而这次是你来分享……你在日常生活或工作中发现 AI 有所助益的方式。有什么工具……有什么 AI 工具你觉得有用愿意分享的吗?
Nabeel S. Qureshi: 哦,天哪,有太多了。我给你举几个例子。我经常用 Wispr Flow。这是一个语音键盘应用,你说话它就会转录。当你用 LLM 快速迭代时特别好用,有时候你需要输入段落那么长的提示词,直接说出来就方便多了。所以 Wispr Flow 我很喜欢。
Lenny Rachitsky: 补充一下这个,你按一下按钮就开始说话——
Nabeel S. Qureshi: 对。
Lenny Rachitsky: 它就会把你说的写出来。这些产品其实已经存在很久了,Dragon Dictate 之类的。现在的区别是这些转录功能真的变得非常非常好用了?
Nabeel S. Qureshi: 我觉得确实是这样。它们用了非常好的模型,所以即使我觉得环境相当有挑战性,它也几乎不会出错。而且用户体验也做得很出色。所以这是个非常好用的工具。
Nabeel S. Qureshi: 我还喜欢用 Claude Code 来开发。虽然我有些抱怨,但它就是有一种让人上瘾的魔力——你可以直接告诉它要做什么。这基本上是一个运行在你电脑终端里的工具,你输入 Claude,就会打开 Claude 界面。设计得很可爱,非常精美,然后你直接告诉它要做什么。它直接在文件系统上操作。所以如果你说”嘿,创建这些文件”,它就会帮你创建好,你不需要自己在 Finder 里折腾。然后它还能处理这些非常复杂的 Pull Request,而且执行得相当好。对我来说,这是一种非常令人兴奋的 AI Agent 预览。
Lenny Rachitsky: 这正是我想问的。所以这本质上就是一个 AI 智能体工程师。我之前不知道 Claude Code 还能做这个。很酷。
Nabeel S. Qureshi: 是的,它算是一个引导式智能体,但确实非常好用。然后我只是很享受……你知道,每周都有新的、很棒的东西可以玩。过去七天我一直在测试 Gemini Pro 2.5。非常出色的模型。我有时候不太喜欢 Google 的用户体验,但我当时就在玩这个。日常生活中我用 LLM 做各种事情。有一天我在报税,需要根据一些元数据对一堆交易进行分类,于是我很快写了个脚本,它就搞定了。
Lenny Rachitsky: 听你描述这些 AI 工具时脸上那个笑容真的很棒。我觉得很多人都是这种感觉,“天哪,我要关注的东西太多了。听到的这些东西,要试的工具太多了”。我就喜欢这种氛围——太不可思议了,太有趣了。我们需要更多这样的心态。
Lenny Rachitsky: 好的。现在进入播客的另一个常设环节。你会经历双重惊喜——反常识角。问题来了:有什么事情是你相信而大多数人都不信的?
反常识角:大学教育的价值
Nabeel S. Qureshi: 我觉得上大学是很棒的事情。在科技行业内部这可能算是有点反常识的观点,在更广泛的经济领域可能不是这样,但我经常看到有人说,“如果你 18 岁就能退学直接开始工作,为什么要上大学?“我觉得这完全错误,不过对于 5% 的人来说这可能是好建议——反正他们本来也会成为你的Fellow。但大学是为数不多的你能结下真正深厚友谊的时期之一。你通常会在一个不错的校园里。如果你在北美,你可以把所有时间都用来思考、写论文、读书、和朋友聚会。
实际上这段时光非常珍贵,在你 21 岁之后很难再找到这样的时间了,因为你要付房租、要工作、要处理各种事情。就算你赚了很多钱、休假一段时间,你的所有朋友也都在工作,你头上总是像悬着一颗定时炸弹。
所以在最开始的那么三四年里,深入钻研各种不同的知识领域、尝试不同的事物、更好地了解自己,这真的很宝贵。我是很坚定的大学支持者。我不会评论 ROI 什么的。我个人认为 ROI 很好,即使美国的学费确实有点高,但这可能是我在科技行业内的反常识观点:不要从大学退学,除非你有很好的理由。
Lenny Rachitsky: 这很有趣,这竟然算是反常识。这听起来确实反常识。我在这里的大学时光很开心。好的。Nabeel,在进入非常精彩的快速问答环节之前,你还有什么想分享给听众的吗?
Nabeel S. Qureshi: 没有了。我觉得现在是世界上一个非常激动人心的时刻。AI 可能让人疲惫,但它真的为以各种方式建设更美好的世界打开了可能性。所以我觉得每隔几个月就重新审视一下自己在做什么,确保它与你认为 AI 的发展方向一致,确保你正在做的事情如果成功的话具有很高的潜力。我认为这比以往任何时候都更重要,因为我们现在拥有的技术杠杆达到了历史最高点。
Lenny Rachitsky: 让我深入追问一下。对于那些想要做到你所说的事情的人来说,有什么能帮助你理解 AI 走向并与之对齐?有什么有用的信息来源和新闻吗?还是就是自己动手玩玩?你的建议是什么?
Nabeel S. Qureshi: 这是个很大的问题。我经常用 X 来跟踪 AI 动态,所以建议你找一个好的 Twitter 列表,也许可以从中关注一些人。有一些不错的 Newsletter。我很喜欢 Latent Space,我知道他的 X 账号是 Swyx,S-W-Y-X,我不记得他真名叫什么,但那个 Newsletter 非常好,而且技术性相当强。如果可以的话,我建议尽量看技术性更强的 Newsletter。现在关于 AI 的哲学讨论或 AI 政策类的内容很多,如果你在这个领域那很好,但这个领域很容易产生很多观点。你读这些不一定能学到很多东西。
但我觉得重要的是要了解正在发生的事情,尽可能经常地重新审视自己的工作流程。确保和你一起来到这里的人会成为那种与 AI 融合的混合赛博格。说到这里,国际象棋领域其实就印证了这一点——如果我可以稍微绕一下路的话——在 2010 年代中期取得最大成功的象棋棋手,其实是最早采用神经网络象棋引擎的那批人。所以当 DeepMind 做出那项成果之后,很快就有了一个叫做 Leela 的开源版本,你会发现像 Magnus Carlson、Fabiano 这样最顶尖的棋手,他们是与 Leela 融合得最深的人,他们学会了它怎么下棋,然后开始模仿它的着法。
Nabeel S. Qureshi: 所以我觉得,尽可能地让自己成为赛博格。然后我觉得这有点像杠铃式的做法——偶尔走出去接触一下大自然也很重要,只是为了保持自己的精神健康。
Lenny Rachitsky: 非常好的建议。那么,Nabeel,我们进入激动人心的快问快答环节。准备好了吗?
Nabeel S. Qureshi: 准备好了。
Lenny Rachitsky: 开始。首先,有没有两三本书是你最常推荐给别人的?
Nabeel S. Qureshi: 脑海中想到的第一本是 Keith Johnstone 的《Impro》。这实际上……我在那篇文章里写过它。这也是我们经常推荐给别人的书之一。我觉得这真的是一本非常有趣的书。名义上,这是一本关于即兴戏剧的书,我相信这个人是这个领域的先驱。他是一位英国人,叫 Keith Johnstone,大概活跃于六七十年代到八十年代。《Impro》是一本关于创造力以及社交行为如何运作的非常有趣的书,基本就是他在即兴戏剧课上教的内容。这是一本非常奇特的书,充满了大量令人难以置信的奇怪想法。书中有很多非常有战术性的东西,比如在第一章他会告诉你做一些事情,就是为了打破你自己的思维框架,非常疯狂的东西。他会告诉你倒着走路,一边从一百倒数,一边思考你正在纠结的某个问题,有各种这类奇怪的事情。但我发现这本书每一页的 Ideas 密度极高。关于社交互动如何运作、以及地位等因素如何影响你的社交行为的这些概念非常重要。我认为这就是为什么我让每个前线部署工程师都读这本书,原因很简单——我觉得它能帮助你更好地读懂人、与他们更好地互动,并且更有意识地觉察自己给他人的印象,然后相应地调整自己的表现。
Lenny Rachitsky: 再说一下书名?
Nabeel S. Qureshi: 《Impro》。
Lenny Rachitsky: 《Impro》。好的,我们会在节目简介里附上链接。
Nabeel S. Qureshi: 好的,《Impro》是第一本。我觉得如果要稍微高雅一点的话,也许是莎士比亚的历史剧,有一套叫”亨利四部曲”的,包括《亨利四世》《亨利五世》《亨利六世》。我发现大多数人不读这些,他们读的是《哈姆雷特》或《麦克白》之类的。但亨利系列绝对令人叹为观止。你不需要对英国君主制或英国历史感兴趣才能欣赏它们。它们实际上是我读过的关于权力、权力如何运作、政治,以及如果你想成为一个成功的国王可能需要做出哪些牺牲的最有趣、最有洞察力的书之一。但这可以迁移到其他地方。我觉得在当今世界,一切都围绕着这些突出的人物和个性组织起来,想想这一点真的值得深思。比如你想到现在的政府,你会想到特朗普或马斯克;你想到 AI,你会想到 Sam 和 Dario,对吧?所以我认为重要的是要理解如何思考这些人物的个性,以及他们玩的到底是什么样的游戏。而《亨利四部曲》实际上是围绕这一点的一套非常棒的书。
它们也很容易读——听起来很好笑,但你可以一天读完一部莎士比亚戏剧。大概只有五十页左右,没那么可怕。你只需要适应一下语言。但是,我强烈推荐这本书。我猜你问的是两到三本。我很喜欢 Andy Grove 的《High Output Management》。我觉得这是一本很棒的商业书籍,人们往往在网上看摘要比真正读原书更多。但原书有很多非常有趣的故事和解释……我觉得这本书最强大的地方实际上是 Andy Grove 的思维方式,而不是其中的具体策略。除非你读到他如何得出这些结论,否则你是无法获得这些的。
Lenny Rachitsky: 你推荐的前两本书和大多数人的推荐非常不同,第三本书则是这个播客上被推荐最多的书。我喜欢我们刚才走过的这个光谱。完美。下一个问题。你最近有没有特别喜欢的电影或电视节目?
Nabeel S. Qureshi: 我最近真正喜欢的一部电影是《分手的决心》。这是一部韩国电影,导演是《老男孩》的导演,可能有人听说过。这是一部很棒的电影。我觉得是几年前上映的,基本剧情是:一名侦探在调查一名被指控杀害丈夫的女人,他渐渐开始爱上她,而这开始以各种方式影响他的判断。一部非常有趣的心理学惊悚片,带有一点浪漫元素。视觉上非常美。我觉得现在很多最有趣的电影实际上都来自海外——东亚、南亚,类似的地方。电视节目我看得不多,已经有一段时间了。
Lenny Rachitsky: 作为创始人完全可以理解。下一个问题。你有没有最近发现特别喜欢的产品?可以是应用,也可以是实物,比如水壶。
Nabeel S. Qureshi: 这个问题我没有什么好答案。我觉得自己买的东西不够多。
Lenny Rachitsky: 完全接受。快问快答没有错误答案。继续,有没有对你在工作或生活中经常有用的人生信条,你会用来和朋友或家人分享的那种?
Nabeel S. Qureshi: 有一位叫 Christopher Alexander 的建筑师,他写了一些关于美的非常美丽的书,内容远超建筑范畴。他在 UC 伯克利教书,学生们总是交出平平无奇的作品,这让他非常沮丧。所以他每周都会告诉他们:想象法国有一座叫做沙特尔的大教堂。你必须以沙特尔为目标。你必须做出比那更好的东西。这应该是你的目标,而不是交出你觉得”够好了”的东西。你实际上必须努力比有史以来最优秀的人做得更好。我发现自己经常对自己重复这句话。就是以沙特尔为目标,真的努力去做到。否则,你很容易不自觉地锚定在中间某个位置。而这种锚定你是无意识地在做的。
Lenny Rachitsky: 那这就是你的信条了——以沙特尔为目标?
Nabeel S. Qureshi: 对,对,对。
Lenny Rachitsky: 我喜欢这个。大多数人完全不知道那是什么,但有了这个背景就非常有力量。最后一个问题,你觉得对产品构建者来说最有价值的经典小说是哪一本?
Nabeel S. Qureshi: 我最喜欢的小说是《安娜·卡列尼娜》,我会建议每个人都去读一读。
Lenny Rachitsky: 我现在正在读。我以前从来没读过。
Nabeel S. Qureshi: 不会的。是的,这是列夫·托尔斯泰的作品。这是一部宏大的十九世纪俄罗斯小说,描绘了社会各阶层的一组人物。我认为它非同凡响,因为他最令人惊叹之处在于能够设身处地想象任何人的内心世界。比如……他会短暂地进入某个角色的大脑,也许是把餐食端上桌的仆人之类的,他能写出一整页关于那个人在想什么,然后迅速转回主人公的视角。我认为这是我见过的对这种技能最令人叹服的展示。
我认为,联系到你的问题,如果你想真正做好产品,就必须真正地设身处地站在他人的角度思考,必须真正以他们看待问题的方式去理解。尤其是作为创始人或产品人,很容易陷入自己看问题的方式无法自拔,对吧?你写了一份文档,做了这些标注,你觉得这会很棒。但当你把它拿给某人时,他们并不那么在意。你真的需要锻炼自己的共情能力,理解他们为什么那样看问题,以及他们真正在意的是什么。
Lenny Rachitsky: 这个比喻太精妙了,把所有东西都串联起来了。我也顺便说一下,读书的时候有个小技巧……人们常说让 ChatGPT 的语音模式陪在你身边,我发现读这本书时特别有用,就是直接问这到底是什么意思?书里有很多俄罗斯的舞蹈、舞会和礼仪,你直接问,就像在问”我在读《安娜·卡列尼娜》,这是什么意思?“它就会告诉你。
Nabeel S. Qureshi: 是的。
Lenny Rachitsky: 这是 AI 的另一个很酷的用法。好的。纳比尔,这次对话太精彩了。最后两个问题,以防有人想找你。你在网上哪里可以找到你?听众可以怎么帮助你?
Nabeel S. Qureshi: 在网上找到我的话,我的网站是 nabeelqu.co,我的 X 账号是 Nabeel QU,我最活跃的平台可能就是那里。我的网站上有所有的链接,还有很多文章和有趣的内容。你们能怎么帮我?我建议给我发邮件。我的邮箱在网站上。介绍一下你自己,打个招呼。我很喜欢认识新朋友。现在我不太有空参加咖啡会议之类的事,但我真的从收到有趣的人的邮件中获得很多能量,所以请尽管联系我。
Lenny Rachitsky: 太棒了。大家一定要去看看《纳比尔的原则》。是这个帖子的名字吗?
Nabeel S. Qureshi: 是的。
Lenny Rachitsky: 好的。那是一个很好的起点,然后还有我们刚才聊到的那个 Palantir 的帖子。好的,纳比尔,非常感谢你来参加节目。
Nabeel S. Qureshi: 谢谢你。也很感谢你,伦尼。
Lenny Rachitsky: 大家再见。非常感谢你们的收听。如果觉得这个节目有帮助,可以在 Apple Podcasts、Spotify 或你喜欢的播客应用上订阅。也请考虑给我们评分或留下评论,这真的能帮助其他听众找到这个播客。你可以在 lennyspodcast.com 找到所有往期节目或了解更多关于这个节目的信息。我们下期见。
术语表
| 原文 | 中文 |
|---|---|
| AIP | AIP |
| Andrew Luo | Andrew Luo |
| Anthropic | Anthropic |
| AR-Labs | AR-Labs |
| Asana | Asana |
| Barry McCardel | Barry McCardel |
| BD (Business Development) | 业务拓展 |
| bull thesis | 看多逻辑 |
| Charge | 沙特尔(法国沙特尔大教堂,Cathédrale Notre-Dame de Chartres) |
| Christopher Alexander | Christopher Alexander |
| CMO (Chief Medical Officer) | CMO |
| COO (Chief Operating Officer) | COO |
| Databricks | Databricks |
| David Hsu | David Hsu |
| Decision to Leave | 《分手的决心》 |
| EOP | EOP(评估与选择流程) |
| forward deployed engineer | 前线部署工程师 |
| Foundry | Foundry |
| GoCardless | GoCardless |
| Goodhart’s law | 古德哈特定律 |
| Google Docs syndrome | Google Docs 综合征 |
| Gotham | Gotham |
| Habermas | 哈贝马斯 |
| Head of Product | 产品负责人 |
| Henriad | 亨利四部曲(Henriad) |
| Hex | Hex |
| High Output Management | 《High Output Management》 |
| Impro | 《Impro》 |
| Intercom | Intercom |
| IPO (Initial Public Offering) | IPO |
| Leela | Leela(开源神经网络象棋引擎) |
| Lenny Rachitsky | Lenny Rachitsky |
| local maximum | 局部最优解 |
| Looker | Looker |
| Max Levchin | Max Levchin |
| Mercatus Center | Mercatus Center |
| murder board | 质询板 |
| National Institute of Health (NIH) | 国立卫生研究院 |
| North Star | 北极星 |
| OneSchema | OneSchema |
| Ontology | 本体论 |
| Operation Warp Speed | ”曲速行动”(Operation Warp Speed) |
| outbound email campaign | 主动邮件推广活动 |
| PayPal | PayPal |
| PD (Product Development) | 产品开发 |
| Peter Thiel | 彼得·蒂尔 |
| PM (Product Manager) | 产品经理 |
| product leverage | 产品杠杆 |
| Ramp | Ramp |
| Retool | Retool |
| revenue per engineer | 人均创收 |
| SaaS (Software as a Service) | SaaS |
| SAP | SAP |
| short form video | 短视频 |
| Shyam Sankar | Shyam Sankar |
| Snowflake | Snowflake |
| solutions architect | 解决方案架构师 |
| SVP (Senior Vice President) | 高级副总裁 |
| Ted Mabrey | Ted Mabrey |
| ticket size | 客单价 |
| Tyler Cowen | Tyler Cowen |
| unicorn | 独角兽 |
| Vanta | Vanta |
| Watershed | Watershed |
| YC (Y Combinator) | YC |
| Zero to One | 《从 0 到 1》 |
此文档由 AI 分片翻译(translate_long_document)