Article One is a mechanism design studio. We are starting with the US government and state capacity. Not because the government has unique need for better systems, although they can use careful assistance. But really, because the government knows more about system and mechanism design than anyone else.
It’s very common to say that you should build tech for DC because the government is so behind, because the government needs support, and because the government needs entrepreneurs. This is accurate, but false. Tech needs the government far more than they need us, because they have an essential knowledge base we need to deal with the insanity of managing AI systems.
I worked in Silicon Valley on the AGI lab side before jumping to the US Congress. I learned how wild artificial intelligence could be. Even now in 2026, in the software engineering world, we are still catching on the true potential of existing Artificial Intelligence (AI) systems to our work. The power of loop engineering to continuously execute a process and do valuable work is beautiful and intuitive.
But it’s quite difficult to implement. There are now two kinds of developers: those who are still mourning their craft, and those who have found some satisfaction in becoming managers of AI systems. I feel for both. First we prompted AI models to do things. Then we built loops to prompt AI models. Then we built loops to build loops. It’s an endless cycle that we can find value in if we value what we are building, and will never be as incredible as having the technical insights ourselves.
I find some hope in Nabokov. He wrote that the spiral is the spiritualized circle. Very poetic. Loop engineering is the process of making circles. And in the words of Christopher Nolan, your loop either ends having fixed your problem, or it lasts long enough to become a nuisance. You always have to step in. Every process eventually fails the system that built it.
I might have started in tech, but I was forged in DC. Congress helped me understand how we could impose structure on chaos. On the Hill, the systems we built, the incentive structures we created, and the mechanisms we designed were the things that actually had a lasting effect. This is exactly what the world of tech needs.
Engineering is not over quite yet, but what we engineer has changed. We must build institutions that will outlast us, so that we can take on bolder problems. Until now, the purpose of any technical system we have yet built has been to run processes, and for us to then manage that effort. But now, engineering is increasingly about designing processes that manage processes, and further into that recursive loop.
It’s well known that if you’re a builder, your users are geniuses, and your investors are dumb. (I say this with great love to allocators, especially the ones who back me.) People who make money in early stage investing or have impact in ambitious giving know that they’re not the experts in a problem domain. That’s what makes them good allocators. You knowing that you understand the domain better than your investors and worse than your users in turn enables you to be competent.
And if users are geniuses that makes government employees world-class experts on mechanism design. The US federal government is a massive, sprawling entity. It is an extraordinary enterprise. It does massive things. It does, like, a lot. The American govenrment spends more time thinking, designing and living out extremely complex mechanisms than any other institution in the world that I know of.
Government is insane and the processes are weird and the systems are inefficient. It’s also extremely difficult to do better. This is why I’m so excited to work on state capacity. The world we live in right now is created by the incentive structures that govern it. These systems are being torn apart by many forces. We cannot merely build new things to get a better world. We must build new systems.
A little founder energy is useful here. I talk to people who are excited that AI can help improve the work of policymakers. That email can be made better, and networking, and research tasks. They’re not wrong. AI can be used to build many of these kinds of tool, and I’ve been very happy with the results in my own work.
But these folks trying to speed up their work 10x are still missing the point. Artificial intelligence isn’t just a utility that improves software engineering. Artificial intelligence is a technology that has reshaped the nature of industry and commerce. We can’t just build better carriages; we need to rethink policy and service delivery in the state.
Building for the government is an opportunity to work most closely with those who are at the seating mechanism, at the very frontier of building the systems that build systems. And isn’t that what loop engineering is all about? We have AI systems that can do tasks very well. We build loops to do those tasks. We build loops to build loops and so on.
Really, human beings are loops. We take processes and we implement them again and again. And we as loops are not being designed by another human being. Our loops are in response to the incentive structures and mechanisms built into the world we inhabit.
We are now in the place on this journey where we must build grand designs to contain the loops that can be created. And the greatest system designers of all time did not work on search engine ads. They built economic systems and political systems that shaped humanity writ large for millenia.
If you want to change the world, you just need to design the right mechanism. Sarthak@article.one