I built AI for finance teams who don’t trust AI :)

by•

Spent the last year building AI for finance teams — where a wrong number isn’t a bug, it’s an audit finding. I went in obsessed with accuracy.

The surprise: accuracy was never what made people trust it. What did was the opposite instinct — making sure the AI never decides anything. The model proposes, a rule decides, a human signs, every number traces back to its source.

Nobody wants a smarter black box. They want to stay in control.

I might be wrong, though. For anyone who’s shipped AI into something high-stakes — what actually earned the trust: accuracy, or control?

30 views

Add a comment

Replies

Best

Control, for me too. Most of my work runs through agents now, email, support, docs and code, and the rule I keep is that the model doesn't decide. Anything that's logic, like dates, amounts or validation, gets built by a script. When I really do need a model to make a call, a different model checks it.