Articles

Insights and perspectives

How we think about systems, and what that means for the companies we build them for.

We ran one application on seven databases. The benchmark found a bug in our own code.Most database benchmarks measure workloads nobody runs. So we replaced the storage engine underneath a live system seven times, kept the data, the queries and the indexes, and measured what actually happened. The most valuable result was not a timing.

AI didn't replace developers. It replaced the easy part.Code was never the hard part. It was the visible part. Generating it is becoming almost free, and every line of it is still someone's responsibility.

Systems that fail to become platforms will dieSaaS was the right answer while software was expensive to produce. AI removes that constraint — and with it the reason the fixed application should stay the unit of software.

The third software option nobody offered youFor decades software meant choosing between renting a product you never own and owning one that slowly decays. There is a third model: you own the system, and the platform underneath it keeps evolving.

Beyond Jevons ParadoxCheaper AI means more AI. That is the first-order effect. The second is that execution costs shape which ideas we let ourselves consider at all — and those costs just moved.

When everything can be built, the real question is what is worth buildingSoftware has stopped being the limiting factor. What slows an organization down now is not building, but deciding what is worth building.