Yusif Zahmati
Writing

What end-of-day taught me about building software

Draft · 4 min read

My first job in a bank was not a programming job. I was an end-of-day engineer, which means I was the person who stayed behind to make sure the day actually closed: that every transaction landed where it should, that the numbers reconciled, and that tomorrow could start from a clean position.

Nobody notices when end-of-day works. Everybody notices when it doesn't.

Failures are rarely where you're looking

The bugs I chased in those years almost never looked like bugs. A branch had done something in an order nobody had anticipated. A file arrived two minutes late. A value that had been non-null for four years was suddenly null. The code was fine — the assumptions weren't.

Production doesn't care about your happy path.

That's the habit I took with me when I moved to the engineering side of the same bank: before I write the integration, I go and find out what the business does when things go sideways. Not the documented process — the actual one.

Three things I do differently now

  • I write the reconciliation before I write the feature. If I can't prove a run was correct, I haven't finished.
  • I assume every external system will be slow, wrong, or absent at least once, and I decide up front what happens when it is.
  • I keep a way to answer "what happened at 23:47 last Tuesday" without redeploying anything.

The part I didn't expect

Knowing the domain turned out to matter more than knowing the stack. When you've spent three years watching what a bank does at midnight, a requirements document reads differently. You can tell which line is a real constraint and which one is someone's habit. That saves more time than any framework I've picked up since.

I don't miss the hours. I do use the instincts every single week.