Hot Celebrity Orchestration Problem

Term coined by faye-zhang for a class of rigid-orchestration failure:

"Right now Claude code doesn't support a dynamic scaling, meaning you will quickly run into what we call the hot celebrity problem in system design, where one of the agents become incredibly busy while the others sit idle."

A single sub-agent becomes the fan-in/fan-out bottleneck because the orchestration graph is static. Load-balancing requires the main agent to spawn more copies of the hot node on demand, which off-the-shelf Claude-code sub-agent harnesses didn't support as of March 2026.

Why this matters

Explains why a team can parallelise "on paper" and still hit serial throughput. Pairs with harness-engineering and agent-orchestration-stack — the orchestration substrate (not the model) bounds your gains.