Look👀,I keep coming back to one thing about AI agents: can I take the permission back?
Handing an agent access is easy. What worries me is next week, when the task has changed, or I've changed my mind, or something just feels off halfway through and I want it to stop. If undoing that is a hassle, I didn't really delegate. I just crossed my fingers.
This is why Kite has my attention. From what I can tell, the authority stays linked to the user for the whole life of the agent instead of being a single "yes" you give once and forget. That makes more sense to me than trying to predict every situation ahead of time. Things change, and sometimes the smartest move is to pull access before the job is done.
It matters even more once you have more than one agent. I doubt most of us will run one giant do-everything system. It'll be a handful of small ones, each with its own job. Then I'll want to know which agent can do what, who has access to what right now, and whether I can shut one off without breaking the rest.
I'd still love to see how this holds up in real deployments, since diagrams only tell you so much. But I like where it's headed.
My guess is the agents people end up trusting won't be the most capable ones. They'll be the ones that are easy to take things back from.
#KiteBBW $KITE @KITE AI
Handing an agent access is easy. What worries me is next week, when the task has changed, or I've changed my mind, or something just feels off halfway through and I want it to stop. If undoing that is a hassle, I didn't really delegate. I just crossed my fingers.
This is why Kite has my attention. From what I can tell, the authority stays linked to the user for the whole life of the agent instead of being a single "yes" you give once and forget. That makes more sense to me than trying to predict every situation ahead of time. Things change, and sometimes the smartest move is to pull access before the job is done.
It matters even more once you have more than one agent. I doubt most of us will run one giant do-everything system. It'll be a handful of small ones, each with its own job. Then I'll want to know which agent can do what, who has access to what right now, and whether I can shut one off without breaking the rest.
I'd still love to see how this holds up in real deployments, since diagrams only tell you so much. But I like where it's headed.
My guess is the agents people end up trusting won't be the most capable ones. They'll be the ones that are easy to take things back from.
#KiteBBW $KITE @KITE AI