20VC: Cohere Founder on How Cohere Compete with OpenAI and Anthropic $BNs | Why Counties Should Fund Their Own Models & the Need for Model Sovereignty | How Sam Altman Has Done a Disservice to AI with Nick Frosst
Timestamps are as accurate as they can be but may be slightly off. We encourage you to listen to the full context.
Podcast Summary
In this compelling episode, Nick Frost, co-founder of enterprise AI powerhouse Cohere, challenges the AGI hype while making a passionate case for AI's transformative potential in the workplace. From his early days as Jeff Hinton's first hire at Google Brain to building a $6.8 billion enterprise-focused AI company, Frost shares candid insights on why (25:28) he believes current AI technology won't lead to AGI, how Cohere competes against OpenAI's billions by focusing singularly on enterprise tool use (07:43), and why he thinks Sam Altman's predictions about existential AI threats were "academically disingenuous" (57:04). The conversation takes fascinating turns as Frost discusses the future of work, income inequality, and his bold prediction that by 2026, you'll simply tell your computer to "file my expenses" and it will handle everything autonomously (60:27).
Speakers
Nick Frost
Co-founder of Cohere, the enterprise-focused LLM company valued at $6.8 billion with over $900 million raised. Former Google Brain researcher who worked alongside AI pioneer Jeff Hinton, contributing to foundational transformer architecture research before founding one of the world's only foundational model companies focused exclusively on enterprise AI deployment.
Harry Stebbings (Host)
Host of 20 VC podcast, one of the world's leading venture capital and startup podcasts. Known for conducting in-depth interviews with founders, investors, and industry leaders, exploring both strategic business insights and the human elements behind building transformative companies.
Key Takeaways
Challenge the Hype, Focus on Real Value
Don't get caught up in AGI discourse or benchmark gaming—these distractions prevent you from understanding what the technology actually does well. (23:22) The most damaging rhetoric around AI creates existential threat narratives that make it harder to discuss real challenges like income inequality and workforce transitions. Instead, ground yourself in practical applications: ask whether your AI implementation helps people do work they find meaningful while automating tasks they'd rather avoid.
Build Efficient, Purpose-Built Solutions
Success in AI isn't about throwing the most compute at the problem—it's about training models efficiently for specific use cases. (36:04) Cohere trains models to fit on just two GPUs while achieving enterprise-grade performance, spending "orders of magnitude less" than competitors. This efficiency advantage comes from singular focus: training models specifically for enterprise tool use, business data integration, and workplace augmentation rather than trying to be everything to everyone.
Optimize for ROI, Not AGI
The most transformative AI applications exist in the workplace, not personal life. (11:56) Focus on automating the boring, repetitive tasks that employees don't want to do—like expense filing, documentation processing, or API integrations—rather than trying to automate human connection and creativity. This "ROI, not AGI" mindset helps you identify where AI adds genuine value versus where it creates unnecessary friction.
Maintain Technological Realism
Language models are statistical text prediction systems, not digital gods. (25:50) While they generalize remarkably well across tasks, they haven't made independent breakthroughs and won't replace human insight, creativity, and cultural understanding. Understanding these fundamental limitations helps you deploy AI effectively—as augmentation tools that help people focus on high-value work requiring human judgment and interpersonal skills.
Embrace Contrarian Curiosity
Being curious and contrarian is both an asset and a liability in fast-moving industries. (62:22) This trait helps you spot opportunities others miss—like founding an LLM company in 2019 when it wasn't mainstream—but can also lead you astray when conventional wisdom is actually correct. Balance this by staying grounded in customer problems and real-world applications rather than getting lost in theoretical possibilities or industry hype cycles.
Statistics & Facts
- Cohere has hit $100 million in enterprise ARR (00:59), demonstrating significant traction in the enterprise AI market.
- The transformer architecture was invented at Google in 2017 (06:25), establishing the foundation for modern language models, though Google didn't commercialize it quickly.
- Less than 20 companies worldwide are building large language models (07:31) - most in America, a handful in China, Cohere in Canada, and one in France, highlighting the concentrated nature of foundational model development.
