Building AI that Actually works for Finance

In a tech company, AI is the magical pill that delights your customer. Nobody cares how it works.

That thinking breaks completely in finance. At JPMorgan, there is no concept of a black box. No concept of "it works 80% of the time." You're making multi-billion dollar decisions a decimal point change in Excel destroys careers. AI either works 99.99% of the time, or it doesn't work. Full stop.

Most firms importing AI from the tech world haven't made that mindset shift yet. Generic models hit 70% accuracy in controlled environments and fall apart when the stakes are real. Apoorv Saxena made the shift the hard way - building AI at JPMorgan where 95% accurate still means 100% wrong.

Apoorv has built AI inside of three of the most consequential institutions. At Google, working for Fei-Fei Li, launching Cloud AI from scratch. At JPMorgan, where Jamie Dimon personally spent three hours in a meeting questioning first principles - not because he was testing anyone, but because he was genuinely curious about where the technology breaks. At Silver Lake, where he ignored internal operations entirely, went all-in on one portfolio company and proved the model before hiring a single person.

That experience is what Obin AI was built on. This week, Obin came out of stealth with a seed round led by Motive Partners and Fei-Fei Li joining as an advisor - building AI agents operating at 99% accuracy inside some of the world's largest financial institutions, fully auditable, on open architecture, replacing core workflows end-to-end inside their controls.

In this episode of the Modern Capital Podcast, Apoorv and Marc cover:

• What JPMorgan taught him about the difference between AI that delights and AI that can be trusted with a $200M decision
• Why the peanut butter approach (50 use cases, 10% productivity gains, great slide) destroys value instead of creating it
• The Silver Lake playbook: one portfolio company, no hires, prove it works first
• Data readiness: why the buried knowledge in your deal rooms, IC memos and credit agreements has to be unlocked before any agent can work
• The agentic workforce: not tools, not chatbots - agents that understand how your firm makes decisions and manages risk
• Why the firms that win won't be the ones that move fastest, but the ones that build the right foundations in the right order

"When you have transformational technology, incrementalism is a death spiral."

Apoorv has seen AI transformation up close at three of the most consequential institutions in finance, and now he's building for the next one. His framework is clear, his examples are concrete, and the conviction behind Obin is hard-earned rather than theoretical.

This one is worth your full attention.

‍