In a late 2023 experiment, Anthropic set up centralized marketplaces where Claude agents, acting on behalf of employees, successfully executed transactions and demonstrated an ability to understand even unarticulated user preferences.

Anthropic's economics team has shifted its hiring evaluation from assessing a candidate's ability to implement data analysis to their ability to delegate, direct, and evaluate the work of an AI model.

The deployment of AI for complex tasks like biological research or financial modeling is bottlenecked by the need to make large amounts of contextual information available to the model.