Most AI agents today rely on holding wallet private keys or API keys to operate.
This means that "ownership" is directly equivalent to "permission": as long as the agent possesses the key, it can almost do anything.
This model holds true for humans, but for autonomous agents, it is structurally a point of failure.
Keys compress too many different levels of concepts into the same primitive:
- Identity
- Permissions
- Asset control
This design is either-or, making it difficult to accurately define the scope, and revocation is high-risk and costly. Once leaked or misused, damage often occurs immediately and can quickly escalate into systemic issues.
Agents do not need to "own"; what they need is "capability," which is clear authorization for specific actions.
This is exactly the gap that ERC-8004 attempts to fill.
ERC-8004 redefines permissions as delegable and programmable capabilities, rather than possession of keys or assets.
Kite further decouples identity, permissions, and settlement on this basis:
- Permissions can be granted without handing over keys
- Scope, limits, and validity periods of permissions are enforced at the time of execution
- Revoking permissions does not disrupt the system and does not require key rotation
This allows delegation to remain safe and reliable even at machine speed.
Autonomous agents should not "own"; they should be authorized.🪁
