yakthang

If your name isn't David Laing, you probably don't want to use this repository directly. This is a personal workspace for exploring multi-agent orchestration patterns. Rather than cloning it, I recommend pointing your agent at this repo and exploring it for ideas that might work for your own agentic workflows.

What is this?

Yakthang is an experiment in supervised multi-agent software development. It explores the question: what happens when you give a human supervisor a team of AI agents, a shared task board, and a terminal multiplexer?

The name is a play on the Tibetan ཡིག་ཐང་ (yik tang) — the high-altitude plateau where yaks roam free. Here, "yaks" are tasks that need shaving, and "shavers" are autonomous agents that do the work.

Ideas that might be worth stealing

1. The Orchestrator pattern

A long-running AI agent (Yakob) acts as a supervisor — planning work, spawning workers, monitoring progress, and reacting to blockers. The human (mostly) talks to the orchestrator, not the workers directly.

This creates a natural division of labour: the human decides what and when, the orchestrator decides how and who, and the workers just execute.

2. Shared task state as coordination mechanism

All agents (orchestrator and workers) share a DAG-based task board (.yaks/, managed by the yx CLI). Workers read their briefs from it, write progress back to it, and the orchestrator watches it for state changes.