Watching people wrestle a messy data problem, I noticed the instinct is always to think like an engineer: build one deterministic thing that can handle every circumstance, every edge case, perfectly. That’s a lot of work, and it usually breaks the moment reality doesn’t match the spec.
But that’s not what this tech is good at. The strength of AI isn’t perfect rule-following, it’s understanding the core desired state and then using reasoning to bridge the gap with imperfect, incomplete information. You don’t need to anticipate every case up front. You need to be clear about where you’re trying to land and let it reason its way there.
When you hit a hard problem, are you trying to engineer one airtight solution, or could you just describe the outcome you want and let the machine reason toward it?