Latest / The CTO Podcast with Fexingo: Technical Leadership, Architecture, and Engineering Org / How Palantir Builds Software for the US Military
Transcript
- Lucas: This April, Palantir announced a new $115 million contract with the US Army to extend its work on the Tactical Intelligence Targeting Access Node — TITAN. It's the latest in a string of deals that make Palantir arguably the most important defense software company you might not think about day-to-day. Luna: And I think for a lot of people in tech, Palantir is this mysterious, almost controversial name. But what does it actually mean to build software for the military? Lucas: Right. And that's the angle today. Not the politics of defense tech, but the engineering reality. Because Palantir's approach is genuinely different from most commercial software development. They have to operate under constraints that would break a typical SaaS product. Luna: Like what? Give me a concrete example. Lucas: Let's start with the delivery model. Palantir's core platform — Gotham — isn't something you subscribe to. It's deployed into classified networks, often on-premise or in government-controlled cloud environments like AWS GovCloud. Every instance is hardened, audited, and certified to handle classified data up to Top Secret. Luna: So no continuous deployment to production like at a typical startup. Every release goes through a security review that could take months. Lucas: Exactly. And yet the military expects rapid iteration. They want new capabilities delivered in weeks, not years. So Palantir had to build a development pipeline that could reconcile speed with compliance. Their solution was something they call 'continuous authorization' — they never treat security as a gate at the end. It's embedded in every commit. Luna: I've heard that term before. How is it different from DevSecOps in a commercial context? Lucas: In a commercial setting, you might have automated security scanning, but if a vulnerability is found in production, you roll back. In Palantir's world, rollback might not be possible because the system is actively being used in a battlefield planning scenario. So their engineers have to write code that is provably secure from the start — they use formal methods where they can, and every change is peer-reviewed by security engineers who are embedded in the product teams. Luna: That's a huge cultural shift. It means every developer needs a security mindset, not just the dedicated security team. Lucas: Right. And Palantir's hiring reflects that. They look for engineers who have experience with high-stakes systems — people who have worked in finance, aerospace, or even government already. They're not hiring fresh bootcamp grads for the most part. Luna: Let's go back to TITAN for a second. What exactly is Palantir building there? Lucas: TITAN is essentially a mobile intelligence ground station — a truck-mounted system that fuses data from satellites, drones, ground sensors, and other sources into a single picture for battlefield commanders. Palantir's role is the software layer that ingests, fuses, and visualizes that data in near real time. The previous system was built by Raytheon and took years to update. Palantir's claim is that they can deliver new features in weeks because of their platform approach. Luna: So instead of custom-building for every new sensor type, they have a data integration layer — that's Foundry, I think? Lucas: Yes. Foundry is Palantir's commercial platform, but the same architecture underpins Gotham. It's built around an ontology — a semantic data model — that maps all the data sources into a common language. So if the Army adds a new drone feed, Palantir doesn't need to rewrite the visualization. They just map the new data schema into the ontology. Luna: That sounds elegant, but I imagine the ontology itself is incredibly complex to build and maintain. Lucas: It is. And that's where a lot of the engineering effort goes. Palantir has teams of 'forward deployed engineers' — they call them FDEs — who work directly with military users in the field to understand the data flows and map them into the ontology. That's not something you can automate. It requires domain expertise and trust. Luna: So these engineers are literally embedded with soldiers, sometimes in combat zones? Lucas: That's been reported, yes. Palantir has a culture of 'mission focus' that goes beyond the typical tech company. Their engineers sign NDAs that cover classified work, and they're expected to be on call 24/7 if a system goes down. The compensation reflects that — they pay well, but the burnout rate is also higher than industry average. Luna: What about the commercial side? Palantir has been trying to expand beyond defense for years. How does the defense engineering DNA translate to, say, supply chain optimization for a pharmaceutical company? Lucas: It's an interesting question. Commercially, Palantir uses Foundry — the same ontology-based platform — but the sales cycle is very different. Commercial clients don't have the same security clearance requirements, but they do have data privacy and regulatory concerns. One thing Palantir has done is offer Foundry as a SaaS product on AWS and Azure, but their core differentiator is still the integration layer. They claim they can connect data sources faster than a custom ETL pipeline built by the client's own data team. Luna: And does that claim hold up in practice? Lucas: Mixed reviews. Some large enterprises love it — especially in energy and manufacturing where there are many legacy data sources. Others find the ontology too rigid for their specific needs, and the pricing is notoriously opaque. Palantir typically sells multi-year contracts worth tens of millions, so it's not a tool for small companies. Luna: Speaking of pricing — Palantir's stock has been volatile, but they've been profitable for the last few quarters. How much of that is defense vs commercial? Lucas: As of their most recent earnings in May 2026, roughly 60 percent of revenue comes from government contracts, mostly U.S. defense and intelligence. The commercial business is growing, but it's still smaller. And the government contracts tend to be longer-term and more predictable, so they provide a stable base. Luna: But there's also a reputational cost. Some tech workers refuse to work for Palantir because of the defense association. How does that affect their engineering hiring? Lucas: It definitely narrows the pool. But Palantir compensates by targeting a specific profile — engineers who are motivated by mission impact, not just technical challenge. They also have very high standards for coding ability and domain knowledge. The interview process includes a take-home project that's often classified, so candidates can't even talk about it. Luna: That's intense. And it builds a very cohesive culture, I imagine. Lucas: It does. People who join Palantir tend to stay for years, and the alumni network is strong. There's a sense of shared purpose that you don't find at a typical consumer tech company. But it also means they miss out on a lot of talent who would otherwise be a great fit. Luna: So if you're a CTO at a non-defense company, what's the one lesson you could take from Palantir's engineering playbook? Lucas: I think the biggest one is embedding security and compliance into the development process from day one, not as an afterthought. Most commercial companies treat security as a checklist for the release gate. Palantir treats it as a property of every pull request. That level of rigor is expensive, but it pays off in trust and reliability. Luna: And maybe the forward-deployed engineer model — sending developers to sit with users in their environment. That's something a lot of B2B companies could do more of. Lucas: Absolutely. It's hard to scale, but it creates feedback loops that are incredibly tight. Palantir's engineers understand the user's pain because they've lived it. That's hard to replicate with a product manager reading Jira tickets. Luna: Speaking of which — if you've gotten something out of these conversations about how real engineering organizations operate, a small contribution genuinely helps keep the podcast independent and ad-free. We're reader-supported, not investor-supported. A couple of dollars a month makes a real difference for a show like this — buy me a coffee dot com slash fexingo. Lucas: Yeah, it's simple and it goes straight to production costs. And we don't run ads, so this is how the show stays sustainable. Luna: All right — back to Palantir. One thing I find fascinating is how they handle the tension between customization and platform integrity. Every military branch wants slightly different features, but a platform becomes unmanageable if you fork it for every client. Lucas: That's the classic platform vs. product dilemma. Palantir's answer is the ontology. They allow customization through configuration rather than code changes. So the Army can have a different set of data sources and visual dashboards than the Navy, but they're both running the same core platform. The engineering team doesn't have to maintain separate codebases. Luna: But configuration can become as complex as code. How do they prevent the configuration from becoming a mess? Lucas: They have a dedicated team called 'ontology engineering' that manages the semantic models. Those engineers are a bridge between domain experts and software engineers. They write what are essentially declarative rules — 'this data source maps to this object type' — and those rules are version-controlled and tested like code. It's not a silver bullet, but it keeps the platform coherent. Luna: And I assume they have automated testing for the ontology as well? Lucas: They do. When a forward deployed engineer adds a new data source, the system runs a battery of tests to ensure the ontology doesn't break existing queries. It's reminiscent of how database schema migrations are tested, but at a higher level of abstraction. Luna: That's a good analogy. So the ontology is like a living data schema that evolves with the mission. Lucas: Exactly. And that's the core innovation, I think. Most defense contractors build custom software for each contract. Palantir builds a platform that can be reconfigured. That's why they can iterate faster and win contracts like TITAN away from legacy players like Raytheon and Northrop Grumman. Luna: But it also means their lock-in is high. Once a customer's data is modeled in the Palantir ontology, switching costs are enormous. Lucas: That's true. And it's intentional from a business perspective. Palantir's gross margins are around 80 percent, partly because of that stickiness. But from a customer standpoint, it means you have to be all-in on Palantir's way of thinking. Luna: So what's next for them? They've been talking about AI integration — using large language models to help query the ontology in natural language. Lucas: They've had an AI platform called AIP for a couple of years now. It essentially lets users ask questions like 'show me all supply convoys within 50 kilometers of the forward operating base' and get a visualization. The challenge is making sure the AI doesn't hallucinate when lives are on the line. So they've been very careful about grounding the LLM in the ontology and restricting it to only use verified data sources. Luna: That seems like a perfect use case for a constrained domain — you have a well-defined data model, so you can limit the AI's output to only what's in the ontology. Lucas: Right. And that's where I think Palantir has a real advantage over general-purpose AI assistants. Their data is already structured and governed. Adding an NLP layer on top is almost natural. Luna: If they can pull that off, it could be a massive productivity boost for military analysts who are drowning in data. Lucas: And for commercial analysts too. Palantir's Foundry already includes AI features for supply chain and logistics. The same principle applies — ask a question, get an answer from the ontology. It's early, but the trajectory is clear. Luna: So Palantir is essentially betting that the ontology-first approach will win across both government and enterprise. It's a big bet. Lucas: It is. But they've been making that bet for twenty years now, and they're still around, profitable, and growing. That's not nothing.