Comment by monomial

Is this a common pattern to have an agent request a sandbox? I feel like I'd want the whole agent running in it's own sandbox to begin with. Firecracker does look like a decent solution for that.