Building B2B AI-Native Startups: Pre-Seed Insights and Industry Selection
A VC's playbook for B2B AI startups: Why the "5% rule" (AI only needs to beat the worst workers) makes certain industries ripe for disruption, and why MLE talent is now scarcer than opportunities—reversing typical startup dynamics.
Read Original Summary used for search
TLDR
• Sequoia's "5% rule": Industries are ready for AI disruption when agents outperform the bottom 5% of workers—and 66% of enterprise AI pilots now show measurable ROI
• Four conditions for picking industries: APIs for agents to interact with, proper benchmarks, orgs willing to spend ($20M+ on brittle UIPath workflows), and access to cracked MLEs
• Power law of AI spending is extreme: Cursor = 70% of Fireworks revenue; one Mercor customer pays $6M/week
• Tactical move: Raise overvalued seeds to give founding engineers 2-5% equity (vs 10-14% option pools) to compete for scarce MLE talent
• Skip YC's 5-10% equity tax if you have customer access—next mature markets are finance, healthcare, law (domains lacking pre-training data but requiring reasoning + tool calls)
In Detail
The author argues we're in a rare infrastructure shift where AI capabilities have outpaced available talent—MLE expertise is scarcer than opportunities. The core thesis: B2B AI-native startups should target industries where AI agents can outperform the bottom 5% of workers (Sequoia's framework), focusing on white-collar domains with the right structural conditions. These conditions include: (1) APIs and interfaces for agent interaction, not just UI automation, (2) proper evaluation frameworks to benchmark agent vs human productivity, (3) organizations already spending heavily (legacy enterprises spend $20M+ on brittle RPA tools like UIPath), and (4) access to top MLE talent through pedigreed networks.
The tactical playbook is specific: Customer concentration in AI follows extreme power laws—Cursor represents 70% of Fireworks' revenue, and Mercor has one customer paying $6M weekly. This means founders need direct access to hyperspender enterprises, not just any customers. To attract scarce MLE talent, founders are raising overvalued seeds and allocating 2-5% equity to founding engineers (versus traditional 10-14% option pools). The author argues founders should skip accelerators like YC (which take 5-10% equity) if they can secure customer access and hiring networks independently.
The next mature markets are finance, healthcare, and law—domains that lack copious pre-training data, require highly skilled expertise locked in enterprise workflows, and demand a blend of reasoning and complex tool calls. OpenAI's Grove program and consulting push signal labs are moving into these verticals. The gap between passing a benchmark and performing the actual task remains huge, creating defensible opportunities for vertical-specific teams who can execute services-first engagements that gradually productize.