Isolating AI agents proves safer than isolating code tools

Original: if you run agents that execute arbitrary code: do you isolate the tool or isolate the agent? we tried both.

Source: x.com ↗