Skip to content
3 min read

Why I'm building for defense

I'm pointing IntMaxx at defense and aerospace. Here is why: a belief in America, what I learned from Palantir (where I interviewed) and Anduril, and why sovereign AI fits that work better than anything else.

  • defense
  • intmaxx
  • essay

I'm building IntMaxx for defense and aerospace. I want to say that plainly, and then explain why, and how I got there.

The belief underneath

I believe in America. Not as a slogan, but as a decision about where my work goes. I'm a student at Berkeley, in the gravity well of Silicon Valley, and I've watched a lot of very capable people point their talent at problems that are, honestly, small. I wanted what I build to matter to the country I live in, and I wanted to put that first.

The AI era makes it concrete. The most capable intelligence ever built is being rented out by a handful of companies. If it's going to sit inside the systems a country depends on, I'd rather it be built by people who answer to that country, run on its own machines, and be owned by the people who use it. That's the argument for sovereign AI, and no sector needs it more than defense.

Palantir

Palantir showed me what it looks like to build software for the hardest customers, and to sit next to them while you do it. The forward-deployed engineer, someone who goes to the customer, learns the real problem and builds against it, is the model behind how I think about IntMaxx. I borrowed the term on purpose.

I interviewed with Palantir, and that made the model concrete for me. Start from the customer's actual constraint, not from the technology. Measure the result in numbers they already track. That's the idea this whole practice runs on.

Anduril

Anduril showed me a second thing. For a long time the assumption was that defense belonged to a small club of established contractors. Anduril, as I understand it, built its own products first and then went to the customer with something working, instead of waiting to be handed a requirements document.

I don't have the capital or the hardware for that. What I have is software, and a constraint that fits it.

How I decided

Three things lined up.

First, the constraint matches the thesis. Defense programs hold data that can't leave, run on networks that can't be assumed, and last for decades. Those are exactly the conditions where renting a model fails and owning one works. Everything I'd already been saying about sovereignty is a requirement there, not a preference.

Second, domain beats scale. A model tuned on a program's own documents and logs can beat a general frontier model on that program's tasks. That's a small-team advantage. I don't have to out-spend anyone on training. I have to understand the customer's corpus better than anyone else does.

Third, the customer. People who are held to evidence, who care about getting it right, and whose work matters. That's who I want to build for.

What I'm not saying

I'm early. IntMaxx started in July 2026, and nothing here is a boast. I'm not claiming anything I can't back up. Accreditation, clearances and export decisions stay with the customer's own security people, and I build to fit inside them.

The work I mean is decision support: reading, diagnosing, tracing, drafting, so a person can judge faster. A person decides.

What's next

The Defense page lays out how it works. If you work in defense or aerospace and you have a constraint that rules out every hosted option, I'd like to hear about it.

Alex