Build with AI and trust what you make. Free guide, no signup

Ways to decide what an agent may do

Once there are no screens, the rules the screens were holding have to be written down and put somewhere the agent cannot go around.

When an application had screens, the screens decided what anybody could do. Take the screens away, put an agent in front of the same system, and that decision has to be made somewhere else and written down. This group is about where policy lives once the interface stops holding it: what actually changes, where a rule can be applied, what the old permissions model was never built to answer, and how to avoid rebuilding a model somebody else already maintains.

Say it another way

What is policy, in plain words? It is the answer to who is allowed to do what, and when. In an ordinary application nobody had to write that down, because the screens answered it for them. You could only do what there was a button for. An agent has no screens. It can call anything it has been handed, in any order, as fast as it likes. So the rules that used to sit quietly inside the interface have to be said out loud, and put somewhere the agent cannot go around.

BEFOREa personthe screensevery rule lived in herethe recordsNOWan agentnothing, unless you put it heredeclared, and hard to go aroundthe records
The screens were doing the enforcing, quietly, as a side effect of being screens. Take them away and the same job has to be done by something you wrote down on purpose.

The pages in this group