Everything I Didn't Build
For a few weeks I ran a proper multi-agent orchestrator. The concept was right, and I still think the people building those tools are pointed the correct way. But the bill was absurd, and it took me a while to work out why. It wasn't the coding. It was the talking about the coding. Every turn, agents re-read the whole board, reconstructed project context, decided what to claim, checked…
In an effort to build a more efficient agent orchestrator, the author initially created a sophisticated system featuring an agent registry, scheduler, leases and expiry, JSON tickets, SQLite database, status fields, presence indicators, custom columns, and heartbeat notifications. However, after running the system for a few weeks, the author discovered that the inefficiency lied not in the coding aspect, but rather in the extensive communication required to manage the coding process.
The agent orchestrator spent most of its resources deciding who should perform the work, rather than executing the tasks themselves. In response, the author rebuilt their system using a more streamlined approach, with directories and tickets represented as markdown files, Git for history, and an agent board utilizing directories and markdown files to manage tasks.
Key improvements included an agent registry, scheduler, leases and expiry, JSON tickets, SQLite database, status fields, presence indicators, and custom columns. The author emphasizes the importance of understanding whether agents even know about the coordination process before adding additional machinery.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
