AGENT HARNESS BELONGS OUTSIDE SANDBOX

A new architectural approach suggests AI agent control systems should operate outside isolated sandbox environments rather than within them. The proposal has sparked significant discussion in developer communities about security and flexibility trade-offs.