Recursive Self-Improvement, Rushed and Vibe-Coded? Models as a Division of Labor

The Cognitive Revolution · youtube · 2026-08-28

This Cognitive Revolution AI:AM highlights episode revisits conversations with Louis Kirsch and Damon Falck of Inherent Laboratories, Vercel CTO Malte Ubl, Genesis Molecular AI CTO Sergey Edunov, Arm's Mohamed Awad, and others. The central thread: the meaningful unit of AI work is becoming a division of labor between models—from Inherent's 27B "scientist" model delegating implementation to GPT-5.5 Codex, to production pipelines routing taste, execution, security, and hardware coordination across systems.

The discussion covers implications for recursive self-improvement, verifying discoveries humans may not be able to check, model refusals in security work, and infrastructure for agents acting in the world. Its sharpest stake: whether frontier labs are scaling RL and agentic workflows atop reward environments and vendor pipelines that are too rushed, noisy, or gameable to support the institutional self-improvement they pursue.

Original post →

More from AGI Musings

AGI Musings channel →