20VC: Open Models vs Frontier Models: Who Actually Wins? | The...

Full Title

20VC: Open Models vs Frontier Models: Who Actually Wins? | The $100,000 Token Budget Every Engineer Will Need | Why Forward-Deployed Engineers Are the Future of Enterprise AI with Clay Bavor, Co-Founder of Sierra

Summary

This episode features Clay Bavor, co-founder of Sierra, discussing the future of AI models, the economics of AI, and the operational strategies of a rapidly growing enterprise AI company.

Key topics include the interplay between open-source and frontier AI models, the cost of AI compute, the evolution of enterprise AI adoption, and the importance of company culture and leadership.

Key Points

• Clay Bavor left Google after 18 years to co-found Sierra, driven by the opportunity in language models and agents during a period of technological upheaval.
• Sierra deliberately chose not to train its own foundational models, opting instead to fine-tune existing open-weight models to manage capital expenditure while still building proprietary AI architectures.
• The demand for frontier-level intelligence is considered unbounded, particularly in complex domains like science and law, and will likely continue to drive the development of frontier models despite the increasing capability of open-weight models.
• Token economics are complex, influenced by hardware advancements, the migration of workloads to open-weight models, and the fundamental constraint of GPU availability, which acts as a price floor.
• The future of enterprise AI will involve a mix of open and frontier models, with companies adopting a strategy of using the most cost-effective and capable model for specific tasks.
• Locally running AI models on personal devices is seen as beneficial for some consumer applications but unlikely to alleviate the significant compute demands of large-scale enterprise AI training and inference.
• The US open-weights AI ecosystem may be lagging behind due to a perceived reluctance of major US labs to compete with their own frontier models by releasing highly capable open-weights alternatives, unlike Chinese companies which appear more focused on distilling and offering advanced models.
• Sierra's success in serving large enterprises is attributed to a "forward-deployed engineer" (FDE) model, inspired by Palantir, where engineers work closely with clients to ensure successful AI implementation and adoption, though it's not strictly binary for sales.
• The future of teams will likely be leaner and more high-leverage due to AI, but serving large, complex enterprises like those Sierra works with still requires significant human expertise in understanding customer needs, integrating systems, and building trust.
• Sierra utilizes internal AI agents, like "Pinecone," to significantly boost employee productivity across various functions, from engineering to hiring, streamlining operations and enhancing decision-making.
• The company culture at Sierra is built on three core values: craftsmanship, intensity, and family, emphasizing excellence in execution, a high-impact work ethic, and a balance that respects personal life.
• The hiring process has been significantly adapted to embrace AI, with interviews now often including tasks where candidates use AI tools to build applications, reflecting an AI-native approach to talent acquisition.
• The increasing use of AI in code generation is seen as a potential boon for cybersecurity, as AI can also be leveraged to identify and mitigate vulnerabilities, though it also presents new challenges.
• Founders Clay Bavor and Brett Adcock have a "major/minor" division of responsibilities, with Bavor focusing on company operations and Adcock on sales and engineering, both deeply trusting each other's expertise.
• The company's board meetings are held every six weeks rather than quarterly to adapt to the rapid pace of AI development, and they utilize detailed board memos instead of decks to foster deeper engagement.
• The perceived "unbounded demand" for AI solutions is a unique market condition, driving rapid adoption and making this a crucial time for companies in the AI space.
• The high token spend for top engineers ($100,000+ annually) is seen as an indicator of AI adoption and productivity gains, with potential for this to become a normalized part of engineering budgets.
• A key piece of advice for young people entering the workforce is to master AI tools, as this provides an "unfair advantage" and high value in the current job market.
• The importance of in-person work for building culture, camaraderie, and facilitating apprenticeship and mentorship in a young company is strongly emphasized.
• The founders' values of "craftsmanship, intensity, and family" are seen as critical for building an enduring and excellent company.

Conclusion

The rapid advancement of AI is creating unbounded demand for intelligence, influencing both open and frontier models and fundamentally changing how companies operate and build products.

Companies that embrace AI, invest deeply in their technology, and foster strong cultures of craftsmanship, intensity, and family are best positioned to navigate and succeed in this transformative era.

The future of work will be heavily shaped by AI, necessitating a focus on adaptability, continuous learning, and leveraging AI tools to enhance human capabilities rather than replace them.

Discussion Topics

• How do companies balance the cost of developing or utilizing frontier AI models versus leveraging more accessible open-weight models for specific business needs?
• What are the ethical considerations and potential long-term impacts of highly capable AI agents performing complex tasks traditionally handled by humans?
• As AI continues to automate and augment human capabilities, what skills and mindsets will be most critical for individuals to cultivate to thrive in the future workforce?

Key Terms

Frontier Models

Highly advanced, state-of-the-art AI models that represent the cutting edge of performance and capability, often developed by large research labs.

Open-Weight Models

AI models whose parameters (weights) are publicly released, allowing for broader access, fine-tuning, and use by the community.

Scaled Distillation

A process where a smaller, more efficient model is trained to mimic the behavior of a larger, more complex "frontier" model, making advanced AI capabilities more accessible.

Token Economics

The economic principles governing the use and cost of "tokens" (pieces of text processed by AI models) in AI language models, influencing pricing and usage patterns.

Compute

The processing power required to train and run AI models, typically involving significant hardware resources like GPUs.

FDE (Forward-Deployed Engineer)

A model where engineers are embedded within client organizations to directly implement and support technology solutions.

ARR (Annual Recurring Revenue)

A metric used to track the predictable revenue a company expects to receive from its customers over a year.

MCP Gateway

A system that aggregates main company systems and services, allowing agents to access them.

Agent Architecture

The underlying design and framework for building and operating AI agents that can perform tasks and interact with systems.

OMOTE NASHI

A Japanese cultural concept of selfless hospitality, implying deep care and attention to customer experience.

AI-Pilled

A term used to describe individuals who have become highly enthusiastic and knowledgeable about AI.

Timeline