Autonomy for agentic workflows has nothing to do with "replying more like a person", you have to refine the model for it quite specifically. All the large players are trying to do that, it's not really specific to Anthropic. It may be true however that their higher focus on a "Constitutional AI"/RLAIF approach makes it a bit easier to align the model to desirable outcomes when acting agentically.