Radar ·
AI R&D agents rarely revisit training strategy
A study of 1,338 post-training trajectories reports that the overall strategy changed in only 74 of 3,557 adjacent experiments.
Published on August 22, 2026, the study examined 1,338 post-training trajectories. The overall strategy changed in 74 of 3,557 adjacent experiments, indicating that agents rarely revisited it.
Diaries, skill libraries, and evaluators improved benchmark results but did not prompt strategy changes. The authors suggest that agent workflows may need explicit triggers to reconsider the overall strategy, rather than merely refine its implementation. Consult the original study to verify its methods, data, and conclusions.
Original source: Agentes de P&D em IA raramente revisam sua estratégia de treinamento