“According to the UK AI Security Institute, advances in agent scaffolding provide a performance advantage of only a few months before a new, more capable base model achieves the same results without complex tooling.”