Your Agent Needs a Sandbox, Not Your Laptop

I let AI agents write most of my code. I stopped letting them run on the same machine that holds my SSH keys, my .env files and my push rights to main. The agent is not malicious. It is just

I let AI agents write most of my code. I stopped letting them run on the same machine that holds my SSH keys, my .env files and my push rights to main. The agent is not malicious. It is just confident. Every agent I use can run shell commands. That is the whole point. It is also the whole problem, because a shell does not care whether a command came from me, from the model, or from a sentence hidden in a README the model just read. Three things go wrong, over and over: files get deleted, secrets