← All writing

Check the data before you describe the product

The quickest way to write a bug into a product is to write a requirement that describes how you think it behaves today. Every ticket that says "currently, the system does X" is a claim, and if the claim is wrong, the developer will faithfully build the change on top of a fiction.

So before a ticket describes current behaviour, I query the product's own database and look. It takes minutes. Checking the numbers is also what caught a dashboard whose two panels disagreed with each other before the dashboard went out.

The same habit applies to code. At CIRQL, an early-stage app built in Flutter, Dart and Firebase, I read the codebase with AI assistance before scoping features. It cut scoping time by about a working day, and it meant I could argue feasibility from what the code did rather than from what I assumed it did. Product owners do not need to write the code. They do need to know what it does.

The same principle, applied to research rather than software, is why every figure in my dissertation was verified by hand even where AI had done the first pass.

More writing Write to me about this