AI agents are becoming capable of doing much more than answering questions. They can potentially hold permissions, interact with applications, manage resources, and carry out actions on behalf of people or organizations. That makes the question of who controls an AI agent increasingly important.
If an agent is given authority to perform a governance-related action, there should be a clear answer to several basic questions. Who created the agent? Who owns it? What permissions does it have? What information can it access? And what happens if it makes a decision that nobody expected?
I think this becomes particularly interesting in decentralized systems. Governance is ultimately about assigning authority and creating mechanisms for accountability. AI agents introduce another type of actor that could eventually participate in those systems.
One possible approach would be to give agents limited, task-specific authority rather than unrestricted control. Their actions could also be recorded so that the community can review what happened later. Important or irreversible actions could still require human approval.
What level of authority should an AI agent have in a decentralized governance system, and where should human approval remain mandatory?