**Google engineer sparks debate on AI adoption with '20-60-20' law, prompting rebuttal from DeepMind CEO**

A viral post by former Google engineer Steve Yegge claims that Google's AI adoption lags behind peers, citing a '20-60-20' distribution: 20% heavy AI users, 60% using basic tools, and 20% resisting. Yegge criticized Google's internal ban on third-party coding tools like Claude Code, arguing that Google's own Gemini fails to match competitor workflows. DeepMind CEO Demis Hassabis dismissed the post as false and clickbait. Yegge countered with a challenge: prove that half of Google engineers use 4M tokens/day, vs Anthropic's estimated 10-15M per employee.

**Why it matters:** This exchange highlights a structural cross-substrate tension: enterprise AI adoption is not uniform even at frontier AI companies. The '20-60-20' pattern echoes a classic organizational inertia cited in the AMW substrate—the challenge of moving the middle 60% from passive to active AI use. It also reflects the 'token-maxxing' trend (e.g., at Meta, Writer, Sequoia-backed startups) where internal token usage becomes a KPI for transformation. For Google, the gap is exacerbated by its 'innovator's dilemma'—restricting external tools while promoting internal ones like Agent Smith. The debate updates the Frame 3 (incumbent bear) position in AI coding: even a top lab can struggle with cultural adoption.

**Expert take:** The real signal is that AI deployment velocity now differentiates companies. The 'token-maxxing' movement—endorsed by Reid Hoffman and Sequoia's Sonya Huang—argues that heavy token consumption builds organizational fluency. But as capital cycles tighten, the question shifts from raw token volume to conversion efficiency: how much of that consumption drives measurable productivity. The 20-60-20 framework suggests that the median organization will struggle to capture value unless leadership actively reshapes incentives and tooling. Google's internal tension may be a leading indicator of a broader adoption plateau.