I'm finding that the newer GPT models are much more willing to leverage tools/skills than Claude, reducing interventions requesting approval. Just an observation.