Six months ago we made AI use mandatory for every engineer at Ara Pay.
Not encouraged. Mandatory.
So I started asking every candidate the same question: how do you actually use AI day to day?
Most talk about Copilot autocomplete. Some mention generating scaffolding or writing tests faster. That’s fine.
But occasionally someone answers differently.
They talk about managing context windows. About RAG pipelines they designed themselves. About what happens when the model starts confidently hallucinating. About code that made it to production and the edge cases they had to trace back to the model’s output.
They’re not impressed by AI anymore. They’ve been burned by it.
That’s the tell. Not how much someone uses AI, but whether they’ve shipped something that depended on it and had to own what came out the other side.
We’re three people doing the work of ten because we built that leverage into how we operate. The people we hire need to add to it, not just appreciate it from a distance.
When did AI first disappoint you enough that you changed how you use it?
Leave a Reply