← Claude
IndyDevDan
SEE CMUX SOLVE Multi-Agent Orchestration (Claude Code and Pi Agent)
⏱ 30 min video · 3 min read6 Jul 2026Worth watching
TL;DR
IndyDevDan demos CMUX (a Mac terminal multiplexer) as a solution to three core multi-agent orchestration problems: lack of programmatic access, inability to monitor agents, and slow manual agent startup. He walks through live examples using Claude Code, Codex, and Pi agents running models like Opus 4, Minimax M3, and GLM 5.2, showing how CMUX enables scalable, visible, and scriptable agent fleet management.
Key points
1
CMUX solves three multi-agent problems: no programmatic access to agents, inability to monitor/improve agents, and slow manual fleet startup.
2
CMUX is Mac-only; Linux/Windows/WSL users should use tmux as the equivalent alternative.
3
The mental model is: Windows contain Workspaces, which contain Panes — agent teams are organized at the Workspace level with leads on the left and workers on the right.
4
A CMUX skill can be given to any agent, enabling flat bidirectional communication where any agent can prompt any other agent, not just top-down orchestration.
5
The creator uses a three-tier architecture (Orchestrator > Team Leads > Worker Agents) launched via a just file one-liner, giving access to ~5 million tokens of combined context across a team.
Actionable insights
→
Use CMUX Workspaces to separate agent teams by project or function, with a lead agent pane on the left and worker agents on the right for instant visual clarity.
→
Add a CMUX skill to your agents to enable any agent to read any terminal surface and send keystrokes to any other agent, enabling flat peer-to-peer agent communication.
→
Use a just file with a single command (e.g. 'just fast-cc [feature-name]') to instantly boot a full three-tier agent team, eliminating manual startup bottlenecks.
→
For production hotfixes, launch 8+ agents in parallel across workspaces to race toward a solution — different models (Opus, Sonnet, GLM, Minimax) have different strengths and the fastest correct answer wins.
→
Only adopt a new tool when you have a specific problem it solves — do not pick up CMUX until you are actually bottlenecked by agent visibility or launch speed.
Notable quotes
“An agent you can't monitor is an agent you can't improve.”
“If a tool does not have programmatic access, I just completely ignore it. I don't give it another second of my time.”
“We want to rise with the ceiling of agentic engineering, not the floor of vibe coding.”
Worth watching?
✅
Worth watching the full video?
Watch if you are actively running multi-agent setups with Claude Code or similar tools and want a concrete, live-demo walkthrough of how CMUX solves fleet visibility and launch automation — the key concepts are all here but the live terminal footage adds useful context.
Saved you some time? The creator still deserves a like.
Want this for your own channels?
Add the channels you follow. Every new video summarised and in your inbox the moment it drops. From £4/month.
Try it free