devlog 01: The clever version we threw away.

When we started building VANTA — our accounting platform for firms in Saudi Arabia and India — the first plan was, in hindsight, a little in love with itself. Custom everything. Our own models for parts the world already had good answers for. It read beautifully on a whiteboard. Then we did the honest math on what it costs to be responsible for all of that forever, and we threw most of it away.

What replaced it is duller and far stronger. We stood on a proven, battle-tested core for the things that are genuinely solved — the ledger, the accounting primitives, the boring machinery plenty of companies have already hardened — and spent our real invention only where it’s ours to make: the Gulf compliance layer, the e-invoicing and tax rules that have to be exactly right or the product is worthless.

In a system people trust with money, “clever” is a liability. The bar isn’t impressive. The bar is correct — and correct again tomorrow.

A couple of rules fell out of that, and they’ve held. The screen never does the math: type an invoice and you won’t see the interface tally it — every number is computed once, on the server, in the one place that can be audited and defended. And nothing ships half-built — a change to the data and the change to the interface go out together or not at all, so the two can’t quietly drift apart.

Fidelity over scale

We wrote “fidelity over scale” at the top of the project and left it there. It’s a permission slip to not over-engineer — to skip the flexible abstraction we might want in two years, the config option nobody asked for, the clever generalization that turns one clear thing into three vague ones. Most of the engineering we’re proud of in VANTA is engineering we didn’t do. The hard part was never writing the smart version. It was being willing to delete it.