I used to think the session key was the real security boundary. Once the agent had a valid session, that was the authority services would check.
It’s more like a standing power of attorney that can later issue short-term letters of instruction. The short-term letters can be used for specific tasks, but they only exist because the power of attorney allows them. When the power of attorney is revoked, every letter becomes useless, even the ones still in someone’s hand.
Kite works the same way. The user signs a Standing Intent that sets the agent’s overall capabilities, expiration, and high-level limits. From that, the agent can create Delegation Tokens that bind a specific session to a concrete task. The session key can only spend inside that chain. If you revoke or let the Standing Intent expire, every active session and every future delegation for that agent is immediately bounded, no matter how many sessions are open.
The difficulty is in the scoping. Make the Standing Intent too broad and you weaken the value of session isolation. Make it too narrow and the agent keeps coming back for new authorizations, breaking the flow of multi-step work.
If the Standing Intent is the true root of authority, how much of the day-to-day agent economy will still need human re-signing versus living inside long-lived but carefully constrained intents?
#KiteBBW $KITE @KITE AI
It’s more like a standing power of attorney that can later issue short-term letters of instruction. The short-term letters can be used for specific tasks, but they only exist because the power of attorney allows them. When the power of attorney is revoked, every letter becomes useless, even the ones still in someone’s hand.
Kite works the same way. The user signs a Standing Intent that sets the agent’s overall capabilities, expiration, and high-level limits. From that, the agent can create Delegation Tokens that bind a specific session to a concrete task. The session key can only spend inside that chain. If you revoke or let the Standing Intent expire, every active session and every future delegation for that agent is immediately bounded, no matter how many sessions are open.
The difficulty is in the scoping. Make the Standing Intent too broad and you weaken the value of session isolation. Make it too narrow and the agent keeps coming back for new authorizations, breaking the flow of multi-step work.
If the Standing Intent is the true root of authority, how much of the day-to-day agent economy will still need human re-signing versus living inside long-lived but carefully constrained intents?
#KiteBBW $KITE @KITE AI