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

What permissions were never built for

Composition and context. Two questions roles and privileges were never built to answer, and they are why some rules can only be applied at the moment of the call.

Roles and privileges answer one question well: may this person do this thing to this record. There are two questions they were never built for.

Say it another way

A permissions system asks one question: is this person allowed to do this thing? That works while somebody is clicking around a screen, because they do one thing at a time and you can see where they are. An assistant is not like that. It can do three allowed things in a row, faster than anybody could, and the three together add up to something nobody would have allowed. It can also put the answer somewhere it should not go. Neither of those is something you can grant or refuse, because permissions were never built to think about an order or about where the answer lands.

Composition. Read the contact list. Read the deal values. Send an external email. Every one of those is allowed on its own. The sequence is exfiltration. Permissions were written per object and per verb. Never per sequence. Nobody needed that, because a person clicking through screens was never going to do all three in nine seconds.

each one allowed on its ownread the contact list✓ allowedread the deal values✓ allowedsend an external email✓ allowedthe three togethernobody allowed this
Permissions were written per object and per verb. A sequence was never a thing anybody could grant or refuse, because a person clicking through screens was never going to do all three in nine seconds.

Context. The same call is fine in a direct message and not fine posted into a channel with forty people in it. Roles and privileges have no idea where the answer lands. Nor whether a person confirmed it. Nor whether the instruction came from a human at all, or from the contents of a record the agent had just read.

Neither of those is a permissions bug. They are policy questions that nobody had to answer while the interface was answering them quietly.

They are also the reason you cannot push every rule back to setup time. A sequence is only visible while it is happening. A context is only known at the moment of the call. These two stay where they are.

And they stay yours. No platform is going to decide for you whether an answer belongs in a public channel, or whether three allowed calls in a row add up to something you would not have allowed as one.