Compound LLM systems often solve coordination by adding a higher-level LLM. This meta-agent reads workers' outputs, writes the final answer, allocates later calls, and decides when to stop. That pattern is common, but it adds another model to manage the system.
The UnitBoost paper, posted on arXiv, argues for a different approach: a merge operator rather than a model. The title makes the contrast explicit, and the abstract frames the operator as the mechanism for managing the compound system's coordination tasks.
The abstract is brief, so details of the operator's implementation are not yet clear. But the core idea is that coordination in compound LLM systems need not require another reasoning layer. A merge operator could replace the meta-agent's role.