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 centers on Sierra’s approach to building enterprise AI, especially the tradeoffs between frontier models and open-weight models. Clay Bavor argues that demand for high-end intelligence is effectively unbounded, particularly in complex domains like coding, science, law, and product development. The conversation also explores Sierra’s internal use of AI agents, including systems that connect company tools and accelerate decisions across engineering, hiring, and operations. Another major theme is the rising importance of token economics, with the idea that AI usage could become a major line item in engineering and headcount planning. Finally, the episode covers what enterprise AI deployment requires in practice, including forward-deployed engineers, strong culture, and leaders who can think at both strategic and detailed levels.
Key Takeaways
- 1Sierra is betting on open-weight models rather than training a foundation model from scratch.
- 2Demand for frontier-level AI intelligence is likely to keep expanding rather than saturating.
- 3Enterprise AI companies need internal agent infrastructure to operate faster and more intelligently.
- 4Token usage is becoming a serious economic input, not just a backend expense.
- 5Forward-deployed engineers are increasingly important for enterprise AI adoption.
- 6Great companies depend on truth-seeking culture and leaders with wide dynamic range.
Notable Quotes
""We have not yet appreciated the unbounded demand for call it frontier levels of intelligence.""
""We have a set of our own proprietary fine-tune models, but these are fine tunes on top of open-weight models.""
""They estimate they are between three and 20 times more productive in terms of future shift.""
""I have heard and I have observed that top engineers who are really leaning into Claude Codex and so on are spending more than $100,000 on a run rate basis on tokens per year.""
""The answer is with most things, it's like turned out to be some of both.""
""Brett's majors are definitely sales and then engineering. He is really good at selling software. He is really good as a software engineer.""
""His dynamic range in thinking is second to none zoomed all the way out highest level strategy... all the way into the details, the pixels, the drop shadows, the sound, the texture of something.""
""What people underestimate about Google is when you have the alignment of an ambitious enduring mission, incredibly smart people and a culture that values truth and building in service of that mission.""
Episode questions
Why did Sierra decide not to train its own foundation models?
Clay says the capital expense and ongoing compute burden made full pre-training unattractive for a startup. Sierra instead fine-tunes open-weight models and invests deeper in architecture, product, and application layers.
How does Sierra use AI internally to run the company?
The company built an MCP Gateway that connects major internal systems, letting agents access Slack, docs, board materials, and operating reviews. That powers Pinecone, Sierra Brain, and other tools that help with engineering, hiring, and strategy.
Will token costs keep falling over time?
Clay thinks token economics are more complicated than simple price declines. Hardware efficiency may lower costs, but rising demand, reasoning models, and compute scarcity can keep the effective floor high.
Why does Sierra emphasize forward-deployed engineers?
Because enterprise deployment requires understanding customer workflows, systems, and business objectives in depth. Clay says this approach shortens time to value and helps move AI from demo to production faster.