Chapter 8: Teams for AI-First Companies¶
Scale AI didn't grow from a $1 billion valuation to $14 billion by hiring a new workforce1. They transformed the one they had—data labelers becoming RLHF specialists, basic annotation evolving into sophisticated model evaluation.
At Yirifi, just 2 people built 15 backoffice microsites. But those two people aren't doing two jobs each—they're architecting while AI builds. What made that possible wasn't heroic effort or coding speed. It was a specific capability that changes everything about how teams operate.
The temptation is to ask "how do we add AI people?" The better question: "how do we make everyone AI people?" The answer isn't a reorg. It's a transformation that touches hiring, training, culture, and the fundamental definition of what each role does.
"Just 2 people, 15 backoffice microsites. But those 2 people aren't doing 2 jobs each—they're architecting while AI builds. The skill that matters: knowing what to ask AI to do, and recognizing when AI got it wrong."
Universal insight: AI doesn't replace people. It transforms what people do. The skill that matters isn't coding speed—it's knowing what to ask AI to do, and recognizing when AI got it wrong.
Memorable close: "That's not a new role. It's every role, upgraded."
What You'll Learn¶
-
The 4 Team Models: Embedded, platform, hybrid, and everyone-AI—with Uber's 400+ ML projects and Airbnb's 12% order value boost as proof points. 95% of AI pilots fail. Structure determines which 5% you're in.
-
Own Your Domain, Share Your Foundation: The tension that kills AI initiatives—and how Uber runs 5,000+ models because they drew the domain/platform boundary right. OpenAI's billing incident shows what happens when the line isn't clean.
-
Hiring for AI-First: Meta's new AI-enabled coding interview, 71% of leaders preferring AI skills over experience, and why effective developers reject 12% of AI-generated code. The weighting has shifted.
-
The 90-Day AI Fluency Program: Microsoft's 85% completion rates, 300%+ ROI over three years, and why the 8% drop in training investment alongside 130% AI spending increase is the warning sign that separates winners from failures.
-
How Roles Are Changing: Prompt engineers earning $335K at Anthropic, engineers spending 70% on architecture vs 30% on coding (the inverse of five years ago), and why the skills that specialists learn today become universal tomorrow.
-
Building AI Culture: The 70% of failed AI initiatives that fail for culture, not technology. One in three workers sabotaging rollouts. 53% hiding AI usage. And the practices—from prompt libraries to AI hackathons—that separate the 27% doing new work from the rest.
The Real Question¶
The pattern holds across every section: transformation, not replacement.
For startups, this is an opportunity to hire differently from day one. Test for learning velocity instead of framework expertise. Build prompt libraries instead of knowledge silos. Design for everyone-AI before you have the headcount to justify specialists.
For established organizations, the path is harder but the destination is the same. Start with the 90-day fluency program. Pick receptive cultural pockets rather than mandating organization-wide change. Measure productivity gains and share them widely. The transformation takes 18-24 months, but the alternative is watching competitors transform while you debate.