The Hive: Multi-Agent Orchestration

The hive is under construction. This structured content stub gives The Hive a real route, manifest, and companion surface while the lesson text is being written.

The course builds on cmux's programmable terminal control plane to spawn, monitor, steer, and recover fleets of AI agents — all from a single terminal session.

Lesson outline

1. The Control Plane — cmux as agent infrastructure — workspaces, surfaces, refs, and why your terminal multiplexer is the orchestration layer.
2. The Bridge — How pi-cmux turns a terminal multiplexer into an agent-aware control surface — sidebar status, lifecycle hooks, and tool activity.
3. The Fork Bomb — Why spawning agents that spawn agents is the first failure mode — and the three-layer defense that prevents it.
4. Your First Spawn — Build spawn_pi — a single tool call that launches a worker agent in a visible terminal pane you can watch work.
5. The Fleet Lifecycle — Read what workers are doing, steer them mid-flight, and shut them down — the four companion tools that make spawn_pi useful.
6. Shell Escaping Nightmares — When your prompt contains ! and zsh interprets it as history expansion — the real-world debugging story that led to temp files and @file syntax.
7. When Designs Collapse — The hive workspace pattern looked great on paper — dedicated workspace, nested splits, workspace naming. Reality deleted 88 lines.
8. Completion Detection — The gap between 'tools the LLM calls' and 'infrastructure that runs alongside the LLM' — a polling loop that watches workers and notifies the orchestrator.
9. The Attention Model — Two signals for two audiences — sidebar is the quiet cross-workspace broadcast, notifications are the loud focus-stealing alert.
10. Fleet Dashboard — A TUI widget that shows all active agents — status, elapsed time, last activity — without switching workspaces.