A runbook is a documented, step-by-step procedure for a routine or emergency operation, and it is often automated. It records what to do, in what order and who is responsible, for tasks such as restarting a service, rotating a credential or responding to an alert.
It matters because it makes operations repeatable and reduces mistakes under pressure. When an agent or a script carries out a runbook, the permissions it needs can be defined for those steps only, instead of giving broad standing access.
A runbook is only useful if it is accurate and kept up to date, and an old one can cause harm if it describes systems that have changed. Automating one also means that errors are repeated quickly, so it should be tested and have clear limits on what it may do.
To keep them useful, write them for specific scenarios, test them regularly, review them when systems change and give any automated runbook the minimum permissions required for its steps.