This blog post discusses the complexities of sandboxing coding agents, emphasizing that there are distinct considerations for local versus remote sandboxes. It highlights the different threat models and controls needed for each type and warns against conflating the two approaches.