Running AI agents like Claude Code from cloud VMs reveals a fundamental mismatch between Unix's design and modern AI workflows. The standard approach of using tmux to maintain persistent sessions is a workaround for Unix's assumption that interactive processes should die when terminals disconnect. This creates friction for developers who need long-running, stateful conversations with AI agents.

Unix's two-mode system—interactive for humans, daemon for services—works for its original use cases but fails for AI agents. Agents need to be long-running like daemons yet conversational like interactive sessions. They maintain context, survive disconnections, and allow users to reconnect and resume exactly where they left off. But Unix provides no primitive for 'persistent interactive processes,' forcing developers to rely on tmux and similar hacks that paper over this architectural gap.

The solution might already exist in BEAM, the Erlang virtual machine. Designed in the 1980s for telephone switches, BEAM's process model treats interactive as message passing rather than terminal attachment. Each process is lightweight, isolated, addressable, and persistent—exactly what AI agents need. Instead of fighting Unix's assumptions, BEAM provides a foundation where agents can run indefinitely, maintain state, and communicate without caring about terminal connections.