On boring systems

There is software that is expensive because it is exciting, and software that is expensive because it is right. The second kind never makes the demo reel. It just keeps a promise, at 3 a.m., for the thousandth time, without anyone noticing.

I build money software, so I am biased toward the unremarkable. The money path is one place the machine must not improvise: no model in the authorization path, no probabilistic decision at the moment of truth, only rules that can be read, replayed and verified. Novelty is a feature in the product; in the money path it is a liability.

Change that only makes things better

Monotonicity is the discipline I keep returning to: change that only makes things better, never differently. Better latency, better errors, better receipts. If you cannot replay a week of transactions and observe exactly the same decisions, you do not have a system; you have a performance. Retries are the classic test: idempotency is what makes a repeat safe instead of a duplicate.

Boring is not the absence of ambition. It is ambition that learned where the risk is, and moved it somewhere safe.