11. Securing Autonomous Agents (Permissions, Auto Mode and the Sandbox)

You know, usually when you bring on like a brilliant new assistant, there's this expectation of a honeymoon period, right?.

It goes beyond the title with direct discussion of agent, system, rules, including: You sit them down at their desk, you hand them the keys to the filing cabinet, you give them some massively complex project, say, fixing the legacy accounting code base.

1. 01Because if that assistant knocks on your office door every single time they need to, you know, open a drawer or read a file or run a basic calculation, you aren't actually saving a
2. 02We're looking at how you let an AI agent run autonomously without accidentally, you know, destroying your code base, force pushing broken code to production or just completely comp
3. 03So our source material today is this masterful extraction from a brand new release, Chapter 11 of the book Running Claude, The Operator's Guide to Chat, Cowork and Claude Code by S