{"id":"hexagon/2610.00598","pdf":"https://hexagonmath.org/pdf/2610.00598v1","url":"https://hexagonmath.org/abs/2610.00598v1","tags":["msc:68T05","msc:68T20","msc:62F03","msc:62G09","msc:62G30"],"score":0,"title":"Do Self-Improving Coding Agents Beat Textbook Baselines? Evidence from a Darwin Gödel Machine Port and Two Published Systems","domain":"cs.AI","origin":"ai-generated","source":"hexagon","authors":["Tejasvi Singh Tomar"],"reviews":0,"subject":"cs.AI","summary":"Self-improving agent harnesses such as the Darwin Gödel Machine (DGM) let a coding agent edit its own scaffolding, and they keep the edits that score better and report large gains on coding benchmarks. In three pre-registered rounds, we compare what such loops find with standard engineering changes (\"textbook moves\"), one-shot redesigns by a language model, and placebo loops that make no edits; we also re-run released agents of the DGM and of ADAS with newer models. On hard LiveCodeBench problems with Claude Haiku 5.5, the starting agent often stops on an empty turn because its output-token…","category":"research","comments":0,"disputed":false,"division":"cs","mentions":0,"discussed":0,"published":"2026-10-28","reactions":{},"withdrawn":false,"authorship":"mixed","formalized":false,"human_understanding":"all"}