How does Anthropic assign tasks to AI by assigning exclusive roles and requesting weekly review? 】 According to Beating monitoring, Anthropic recently disclosed its engineering experience in operating a human-machine collaboration team for several months. Multiple employees work together with multiple intelligent agents with independent system credentials in Slack. The intelligent agent is directly mounted on the team roster and communication thread, with clear division of labor and autonomous project promotion like human employees. In order to effectively integrate intelligent agents into the team, collaboration defaults to fully disclosing work. Due to the complete reliance of intelligent agents on retrievable text to understand the background, companies will set security boundaries at the workspace level and default to fully open them to intelligent agents, avoiding cumbersome single document authorization decisions. The team assigns proprietary roles to different agents by writing skill files (such as assigning specific agents as software release managers) to prevent employees from running their own personal AI and causing team information fragmentation. The autonomy of intelligent agents is directly proportional to the reliability they exhibit. In specific practice, an engineering supervisor dispatched an intelligent agent to independently fix 500 bugs and required the agent to submit weekly reflection reports containing mistakes and lessons learned to avoid recurrence. To prevent risks, the team uses a Doer verifier mechanism, where one agent reviews the work of another agent. After the intelligent agent gains full trust and operates independently, the team will also train and guide the agent to learn how to save human attention by merging daily questions and setting up workload barriers to ensure the sustainable operation of the human-machine team. [Original link]