Two competing visions shape the next decade of AI. One view sees a small cohort of builders – the emerging technical clergy – controlling frontier models and dictating how society uses them. The other imagines billions of people learning to command autonomous agents, keeping the power distributed.

High‑profile forecasts back the first view. Dario Amodei warned in May 2025 that AI could cut half of entry‑level white‑collar jobs within five years. Sam Altman said July 2025 customer‑support roles would disappear. Mustafa Suleyman predicted full automation of computer‑based professional work by May 2026, while Elon Musk called for universal high income in April 2026.

Frontier breakthroughs illustrate the clergy model. GPT‑5.2 Pro and Harmonic’s Aristotle formally solved Erdős Problem #728 in January 2026. Claude 4.6 discovered 500+ high‑severity vulnerabilities in open‑source code in May 2026. Anthropic’s Mythos Preview found thousands of zero‑days, yet access remains restricted to vetted partners.

Developers experience a widening performance gap. A single engineer using 64 Claude Code instances rewrote a 535,496‑line Zig library to Rust in 11 days, a 100× speedup versus a 1‑year team effort. MET R’s 2026 study showed experienced coders gained only 19 % speedup, while most users stayed at 1×.