Summary
IndyDevDan explores CMUX — a tmux-based terminal multiplexer — as a solution to three persistent multi-agent orchestration problems: lack of programmatic access to agents, inability to monitor multiple agents simultaneously, and the slow manual process of booting agent teams from scratch. With 15 years of engineering experience, the presenter frames CMUX as infrastructure for “agentic engineering” rather than “vibe coding,” arguing that agents you cannot observe are agents you cannot improve.
The video demonstrates a three-tier orchestration architecture where orchestrator agents prompt lead agents, which in turn delegate to specialized expert agents. A 2×2 fleet is assembled live using Claude Code (Opus 4.8), Codex (GPT-5.5), and two Pi agents running Minimax M3 and GLM 5.2 — all assigned in parallel to a security vulnerability audit across a shared repository. CMUX sends notifications to a top-level window when individual agents complete tasks, allowing the engineer to work on separate branches while the fleet runs autonomously.
CMUX is compared favorably to standard tmux for its per-workspace color coding, tabbed identity panels, and reusable session files that eliminate manual team setup costs. The presenter uses his custom Pi coding agent harness (IPI) alongside Claude Code and Codex, treating each model’s distinct strengths as parallel perspectives on the same engineering problem — a concrete illustration of his “scale your compute to scale your impact” thesis. The video also covers how to use AI agents to generate interactive HTML reference documents for learning new tooling.
📺 Source: IndyDevDan · Published July 06, 2026
🏷️ Format: Hands On Build
