**Not everything needs to be a damn AI agent.**
There. I said it. And I will keep saying it.
Warning: hot takes ahead.
Recently, I was teaching a secure AI class, and one of my students was telling me about an AI agent they were building.
1/16
**Not every damn thing needs to be an AI agent!**
There. I said it. ๐ Again.
I've been having conversations with people who are building AI agents and worrying about how to secure them.
And sometimes I find myself wondering...
1/6
I talk about this (and a whole lot more about least privilege and agent permissions) in the latest episode of DevSec Station:
๐ง *What Permissions Should You Give an AI Agent?*
Watch on YouTube or stream/download it on your favourite podcast platform.
5/6
**Telling your AI agent not to do something is NOT the same as preventing it from doing that thing.**
Imagine giving an AI coding agent access to your entire filesystem.
Then you tell it:
"Only modify files inside this project directory."
Very clear instructions, right?
1/5
@florescodes Don't use an agent.. Instead use regular software to do what you need. A lot of the time, that's enough, and it's less risk + easier to control. ๐
Question: **How are you restricting your AI coding agents?** Are you using sandboxing, limited tool permissions, approval gates, or something else?
This is one of the lessons in the latest DevSec Station episode, *What Permissions Should You Give an AI Agent?*
4/5