Chapter Summary: Teams for AI-First Companies¶

Key Takeaways¶

Team structure follows maturity: Start centralized with fewer than 10 AI practitioners. Move toward hybrid past 15-20 people. Everyone-AI only works with high trust, low ego, and comfort with ambiguity.
Domain owns context; platform owns infrastructure: If it requires customer understanding, domain teams own it. If it requires cross-organizational consistency, platform teams own it. Uber runs 400+ ML projects because they got this boundary right.
Hiring has inverted: Systems thinking, evaluation capability, and learning velocity now matter more than syntax memorization. Meta's new AI-enabled coding round tests how candidates work with AI, not whether they can code without it.
AI fluency takes 90 days—but culture takes 18 months: Microsoft achieved 85% training completion rates with structured programs. But 70% of AI failures are cultural, not technical. One in three workers sabotage rollouts. Start in receptive cultural pockets, not organization-wide mandates.
Every role transforms—none disappear: Engineers now spend 70% of time on architecture and validation, 30% on coding—the inverse of five years ago. Prompt engineers earning $335K at Anthropic today teach skills that become universal tomorrow.