A self-optimizing optimizer claims state of the art with no human intervention

Poetiq’s essay argues that recursive self-improvement is the fastest path to superintelligence and, for the first time, describes its approach: the Metasystem is a self-optimizing optimizer in which every problem it optimizes also improves the optimizer itself (Poetiq). What changes each iteration is the reasoning system of harnesses, prompts, code and exploration strategies rather than model weights, and the company claims state of the art on six benchmarks it had not seen, with zero human intervention. It sits in the self-authored-scaffolding line that runs from an agent rewriting its own harness through Prime Agent’s self-rewriting run on ARC-AGI-3, and it makes at the harness layer the claim Weco AI made at the researcher layer three weeks earlier; five days later the harness-over-weights thesis is restated from the other direction when stock Claude Code turns 30% into 96.2% on ARC-AGI-3.