Generative AI could soon create “virtual coworkers” that could help complete complex tasks by organizing specialized agents to analyze data and refine outputs. Senior partner Lareina Yee and coauthors explain that this could be achieved by creating AI-powered agent systems that first understand a natural-language prompt, coordinate specialized agents to fulfill the task, and finally iterate on the output based on user feedback.
Image description:
A flow chart depicts how a generative AI agent system might execute a workflow. There is a series of four steps. In step one, the user provides a natural language prompt to the system, requesting that it complete a task. In step two, the system then interprets the prompt and builds a work plan. A pullout of step two shows that the agent system has a manager agent and three specialist agents: an analyst agent, a checker agent, and a planner agent. The manager agent subdivides the project into tasks and assigns them to the specialist agents. The specialist agents gather and analyze data from multiple sources, collaborating to execute their individual missions. The agent system interacts with external systems, including databases and systems that contain both organizational and external data, to complete the task. In step three, the AI agent team shares the draft output with the user. In step four, the agent team receives feedback from the user, then iterates and refines the output accordingly.
End of image description.
To read the article, see “Why agents are the next frontier of generative AI,” July 24, 2024.