Engineering the organization

I dislike business literature. I see doing business as an art with a value transaction at the core. The fact that something has worked before, or for someone else, does not mean it will work for you. Every domain, every solution, every company is different. There is beauty in bringing your own take to it.

That is how Marek and I see it. We engineered our organization around how we imagine ourselves working every day.

The materialization of that engineering is what we call Atlas, our operations platform. It started as a customer relationship tool. Then it grew to manage our finances. Now it runs most of our operations, minus the pieces that obviously make sense to outsource, like payments and observability infrastructure.

There was a time when outsourcing most of that made sense. But when you do, the connections between pieces of data end up loose. State drifts between services. You paper over the gaps with automation glue like Zapier moving fields around. That model does not make sense anymore. Large language models and coding agents are the perfect tool for turning loose edges into strong database relations across tables. An error is no longer a blob of metadata with a customer attribute buried in it. An error can hold a real database relation to the account that produced it. That account can be linked to a set of service levels, which belong to a service term, which is tied to a document that is semantically searchable. You get the point.

For us, that means we do not need to spin up departments or teams for narrow specialized roles. We can take those roles on ourselves. The organization stays capital efficient, and we keep a level of understanding of the whole system that lets us make better decisions for the business and for our customers.

We could not be more grateful that AI arrived while we were building the company. There was no one to align with except each other, and Marek and I are very aligned here. We are engineers at the core, and we translate the efficiency we want to bring to our users into the way our company operates. That can be frustrating at times, because efficiency is surprisingly rare in organizations. Even something as simple as getting an invoice paid usually means navigating layers of communication.

All of this saves us money, but that is not why we did it. The more capable the models get, and the more granular and correct the context in our system is, the better decisions we can make. The other day we sat down to look at our pricing model. With data from our own system and the market, we produced an iteration of it in an afternoon. That is why people say we ship a lot. We do. We cut the bullshit and engineered the organization to be fast, and that means we lean heavily on AI. Nothing to hide there.

That is why when I hear people call AI useless, or not powerful enough, or scary, I look at ourselves. I look at the work Asmit has done designing and shipping a rebrand. I look at Marek designing the rack to deploy our own infrastructure. I look at Eduardo making global build caching possible. And I think: not for us. AI is what lets us compete with players sitting on hundreds of millions in investment. They are Goliath. We are David.