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

What It Covers

Clay Bavor, co-founder of Sierra (valued at ~$16B, serving 40% of Fortune 50), covers the open vs. frontier model debate, token economics, forward-deployed engineering as an enterprise sales strategy, and how Sierra operates internally—including board cadence, AI-native hiring, and a $100K annual per-engineer token budget trajectory.

Key Questions Answered

• •Token Budget Planning: Top engineers using Claude Code and Codex are spending over $100,000 annually on tokens—a meaningful fraction of engineering salaries. CFOs should begin treating tokens as a headcount line item: salary plus token budget per employee. Bavor predicts token spend will converge closer to 20% of developer salary, not the 3.8% implied by Benioff's $300M Anthropic spend across Salesforce's engineering base.
• •Open vs. Frontier Models: Companies will mix both model types depending on task complexity. Routine tasks like returns processing suit fine-tuned open-weights models. High-stakes domains—legal, coding, materials science—will drive effectively unbounded demand for frontier intelligence. Chinese open-weights models likely derive capability from distilling US frontier models, explaining their performance advantage over domestically built open alternatives.
• •Forward-Deployed Engineering Motion: Sierra embeds engineers directly inside enterprise customers during deployment, enabling companies like Next and Cigna to go live in six to fifty-eight days respectively. This Palantir-inspired model builds deep business understanding, earns trust, and accelerates time-to-value. Bavor considers it the primary driver of Sierra's speed advantage over comparable-vintage competitors in enterprise AI deployment.
• •AI-Native Hiring Process: Sierra replaced traditional engineering interviews with a build-session format: candidates receive a $150 token budget, choose any coding agent, and build a self-selected application. Evaluation covers architecture, systems design, product thinking, and culture fit. Bavor notes that 22–23-year-old AI-native employees rank among Sierra's most productive, and plans to add AI-native components to every interview role within two months.
• •Board Meeting Structure: Sierra runs board meetings every six weeks rather than quarterly, alternating between three-hour and ninety-minute sessions. Meetings use written memos—six to ten pages—sent in advance instead of slide decks, forcing clearer thinking. Memos explicitly document areas of underperformance and missed opportunities, not just wins, which Bavor credits with generating more substantive board engagement and faster course correction.
• •Internal AI Infrastructure: Sierra built an MCP gateway aggregating all company systems—Slack, documents, operating reviews—into a single server accessible via Claude, Codex, or their internal agent called Pinecone. Pinecone includes a skills library, engineering harnesses, and a personal screening tool Bavor uses to pre-review every hire against his specific criteria. A companion tool called Sierra Brain uses board letters and operating reviews as context for strategic reasoning.