Eye On A.I. #321 Nick Frosst: Why Cohere Is Betting on Enterprise AI, Not AGI
14 snips
Feb 17, 2026 
Nick Frosst, Co-founder of Cohere and former Google Brain researcher, builds enterprise-ready large language models for real business use. He discusses why Cohere prioritizes efficient, private, and deployable AI over chasing AGI. Topics include transformers’ scaling, enterprise deployment and security, model efficiency and distillation, agentic systems, evaluation pitfalls, and AI becoming embedded infrastructure.
AI Snips
Chapters
Transcript
Episode notes
Pretraining Drives Capability
- Most core capabilities come from pretraining, so Cohere trains foundational models from scratch.
- Cohere is one of ~10 companies building foundational models focused specifically on enterprise needs.
Keep Data In-Place
- Bring models to customer environments to keep private data secure and meet regulatory needs.
- Deploy on customers' cloud, VPC, or on-prem GPUs to avoid moving sensitive information externally.
RBC Uses Cohere In Production
- Royal Bank of Canada uses Cohere models for company analysis and document work across teams.
- The models help employees do research and reporting faster within RBC's private environment.