Paul Barnhurst and others discuss why AI still needs human oversight. They explain that telling a model not to hallucinate doesn't work, domain experts must verify outputs, and structured practices like prompt design and audit trails are essential for safe AI use.
AI keeps producing convincingly wrong answers. Everybody knows that, especially when the spec is unclear. And you cannot just tell the AI not to hallucinate. That's just not working because its sole purpose is to produce an output. And so the skill is looking at the output and knowing from real domain knowledge whether it's brilliant or plausible but wrong. I'd like to end like, AI can draft, your judgment is what's right.
So you mean I can't just stick in the instructions, don't hallucinate?
You can do it, but according to my experience, it's not really working.
I would say the most important skill is fine thinking and being able to structure the problem, use the right tools without tokenizing, have the right audit trail, and then use it within the appropriate governance and regulatory framework. And there's something called harness, con— harness prompting that's going to be more important. So it started with prompting, which is you tell it what to do, and then context, give AI a sense of what you're trying to do and the context behind it.
Trust and Leadership Lessons for FP&A From 100+ Top Leaders with CFO Jon Laudie
