The document addresses the governance of agents—AI systems that can autonomously achieve goals in the world with minimal or no explicit human instructions.
The document discusses the current state of agent technologies, their potential and limitations, as well as issues related to their reliability and application in complex tasks. Key findings include: modern agents can perform various tasks but face limitations in more complex and prolonged tasks; there are issues with the reliability of agents and their ability to reason and use digital tools; high data processing costs may become a barrier to the widespread adoption of agents; modern agents significantly lag behind humans in performing tasks that require more than an hour.