The Future of Work

The episode covers genuine ground on AI's role in financial planning - hallucinations, prompting techniques, and when humans beat AI - but relies heavily on repetition of core ideas (trust but verify, prompt better, hire humans for complexity) without proportional depth. Many segments retrace earlier points rather than layering new insight.

“It's almost like it's like having a calculator that's right 95% of the time but won't tell you when its 5% is wrong”

“If you ask what an IRA is, very specific, it does really good at that. But if you ask a more complex question, something that's more involved, what were the best songs in 2025, for example. That type of complex question has an ability to just generate more hallucinations”

The core thesis - AI is a tool, use it for education and delegation, hire humans for complex decisions - is well-established in tech circles. The specific prompting tricks (asking for sources, role-playing as experts) are useful but circulated widely. Few genuinely fresh counterintuitive claims emerge; most advice is sensible mainstream AI literacy.

“It turns out if you're older and have more life experience, you can have a huge benefit”

“I don't necessarily believe it's humans versus AI. I think it's humans with AI versus humans without”

Craig Kirkpatrick is a genuine practitioner: 25+ years in financial services, 3,000+ hours hands-on with AI models, cofounder of Optimal Advisor AI, advisory board member at UC Berkeley Extension. He has real skin in the game and operates at the intersection of finance and AI adoption. However, he is not a household name and his company appears to be a training/consulting firm rather than a major AUM operation, limiting top-tier credibility.

“Craig has more than 25 years of experience in the financial services industry. His perspective comes from more than 3,000 hours of working hands on with AI models and over 10,000 prompts used across advisory and asset management use cases”

“We sold my asset management company in 2022”

The episode includes a few concrete examples: a 100-page real estate document, a plumber with $3M revenue, a 1,150-page tax code, and a financial advisor managing $200M across 300 clients. However, most claims lack supporting data: no specific case studies with outcomes, no quantified before/after metrics on productivity gains, no examples of failed AI decisions in financial planning, and vague statistics ('54% think they're good at AI, only 10% are').

“We saw a number of studies out there, maybe 10% or 20% of individuals are leaning into the AI tools”

“Just a study that just came out that 54% of people think they're really good at AI. And then they looked at the actual output, and it turns out probably only 10% are really proficient at it”

The host (Jill Finlayson) asks solid, logical follow-ups: drilling down on what AI actually helps with, asking about warnings signs of bad output, requesting examples of how brainstorming works, probing the emotional and psychological dimensions of the transition. She doesn't challenge Craig's assumptions aggressively, though; questions are exploratory rather than adversarial. Some softball moments (e.g., accepting broad claims about 50 - 70% of advisor work being automatable without pushback on assumptions).

“Where does AI genuinely help people make better decisions? You're saying it helps them define terms. It helps them compare offerings maybe a little bit. Is it actually helping them to make investment decisions?”

“Are there warning signs that maybe something isn't right?”