Tech Disruptors Capital One on Building AI Moats in Banking
7 snips
Mar 17, 2026 Prem Natarajan, Capital One’s Chief Scientist leading enterprise AI, explains why the bank builds its own AI stack and custom weights. He discusses cloud-native architectures that speed experimentation. He covers using proprietary data with open models, cautious AI use in credit, multi-agent workflows like MACA, and AI as a capacity multiplier for engineers and customer support.
AI Snips
Chapters
Transcript
Episode notes
Cloud Native Enables Agentic Workflows
- Being 100% cloud native put Capital One's data next to massive compute, enabling rapid experimentation and API-driven agent actuation.
- Prem explains agents act by invoking APIs and their microservice, API-centric architecture makes agentic workflows practical.
Slope Into High Risk AI Use Cases
- Don't rush sensitive use cases like credit underwriting; slope into lower-risk applications first to learn safeguards.
- Prem recommends starting with customer experience and fraud pilots to gain behavioral insights before moving to underwriting.
Enterprise Structures Make AI Causal
- Enterprise systems already encode much causality, so AI in enterprises can 'read' causality rather than learn it from scratch.
- Prem argues EGI (Enterprise General Intelligence) will emerge first because APIs and architectures expose causal steps clearly.