Permissions
The rules that decide which actions an agent is allowed to take and which it isn’t.
Think of it like
A hotel keycard — it opens your room and the gym, but not the manager’s office.
Example
The agent can read your calendar and draft emails, but it needs a separate permission before it’s allowed to actually send one.
How it actually works
Permissions are the difference between a helpful agent and a dangerous one, because the same system that can send an email can also delete a database. The safe default is least privilege — grant only what the task needs, and require explicit approval for anything irreversible. As agents act more autonomously, permissioning stops being a checkbox and becomes the core safety mechanism.
For product teams
Decide up front what an agent may touch — permissions are where you draw the line between useful and reckless.
For engineers
An access-control layer that gates each action against a policy before it executes.
Related
- Scoped Access — Narrowing permissions to just what’s needed.
- Approval Gate — The pause where a person grants a one-off permission.
Read anything AI without the jargon
Look up any term in plain English, or save terms as you read with the free Chrome extension.
Open DecoderAdd to Chrome