The standard advice for sandboxing AI agents is "use a VM" or "use Docker."
That's terrible advice for most people.
The VM tax
Spinning up a VM for every AI task:
- Takes 30 seconds to boot
- Uses 2-4 GB of RAM just sitting there
- Requires you to set up networking, file sharing, tooling inside the VM
- Is completely disconnected from your actual working environment
Docker is lighter, but you still need to:
- Set up a container image
- Mount volumes carefully (or the agent can't see your files)
- Deal with filesystem permission nonsense
- Learn Docker
Most people just... don't bother. And then they're running the agent on their main machine with access to everything.
The 10-second solution
On Windows, you can map any folder to a drive letter:
subst Z: C:i-workspaces ask-1
That's it. Point your AI tool at Z:\. The agent can do whatever it wants inside that drive. It can't see anything outside of it.
This is a real OS-level boundary. The filesystem enforces it, not the AI.
Why I use ScsSubst
The built-in subst command works, but mappings disappear when you reboot. ScsSubst (https://subst.scsoi.com/) makes them persistent, and gives you a GUI to manage all your mapped drives. $9.99 lifetime, one-time purchase.
No VM overhead. No Docker learning curve. Just a drive letter.
When you actually need a VM
If you're running untrusted code from the internet, sure, use a VM. But if you're running your own AI coding assistant on your own machine, and you just want to make sure it can't accidentally delete your home folder — a mapped drive letter is enough.
It's the difference between a bank vault and a locked filing cabinet. You don't need Fort Knox to keep your files safe from a confused AI.
Top comments (0)