Eval-driven development: the only way we ship agents
We write the test suite before the agent. Here is how a golden set, a grader and a CI gate keep our AI systems honest.
AZ POS is a multi-tenant point-of-sale system: checkout, inventory, purchasing, invoicing and double-entry accounting. It was built by one engineer, with AI coding tools doing a lot of the typing.
Where it helped most
Boilerplate, migrations, test scaffolding and UI variations. Work that is well understood but tedious moved several times faster.
Where it needed a firm hand
Architecture and money. Clean boundaries, tenancy rules and accounting logic were designed by hand and reviewed line by line. AI tools are happy to produce code that compiles and is subtly wrong.
AI makes a good engineer faster. It does not make the decisions for you.