Thomas Ptacek critiques VSCode Remote SSH’s heavy agent model as LLM coding tools push remote development into sandboxes

Thomas Ptacek argues that LLM coding agents are most useful when they can run and iterate inside isolated remote environments, not on a developer’s laptop. He contrasts Emacs Tramp’s lightweight SSH-based remote editing with VSCode Remote SSH, which downloads and runs a substantial remote agent on the target machine.

As AI coding assistants become more autonomous, demand is growing for disposable, secure execution environments that keep agent mistakes away from local systems. VSCode’s approach reflects a broader shift toward agentic development, but also raises concerns about complexity, trust, and what tooling should install or execute remotely.