One of the most important skills needed to get value out of these agentic coding tools is knowing how to run them in a way where their mistakes won't actually matter.
This is non-trivial, and the tools don't do a great deal to help.
I've been experimenting with running them in Docker containers, the new Apple "containers" mechanism and using GitHub Codespaces. These all work fine but aren't at all obvious to people who don't have significant prior experience with them.
Sandbox alone wouldn't have helped here though. The failure (deletion of files) happened inside the project folder.
What it really needs is a (preferably deterministic) way to revert any sequence of changes and get you back to the original state. And big warning messages before it can do anything that doesn't have an associated rollback command.
Granted, that's not entirely sufficient either; rolling back the creation of a security hole doesn't undo whatever information was leaked while it was open.
I kind of wouldn't be surprised if safeguards for this sort of stuff ends up being a larger industry than AI agents themselves. It really requires a whole rethink of how systems are designed, but without it, the value we can get from AI agents will be severely limited.
This is non-trivial, and the tools don't do a great deal to help.
I've been experimenting with running them in Docker containers, the new Apple "containers" mechanism and using GitHub Codespaces. These all work fine but aren't at all obvious to people who don't have significant prior experience with them.