When the coordinator decomposes a task too narrowly, entire domains get missed. Sub-agents can only work on what they’re assigned — if the coordinator defines “creative industries” as only visual arts, music, writing, and film never get researched. The coverage gap originates at decomposition, not execution.
The decomposition bottleneck
Sub-agents are execution engines. They search, analyze, and synthesize within their assigned scope. They don’t question the scope itself. If the coordinator assigns “analyze digital art, graphic design, and photography,” the sub-agents do exactly that — and the final report covers only visual arts.
This makes the coordinator’s decomposition the single point of failure for coverage. A poorly scoped decomposition produces a thorough analysis of the wrong subset.
Symptoms of narrow decomposition
- Final output covers only one aspect of a multi-faceted topic
- Users report “broad but shallow” results — the system goes deep on what it covers but misses entire areas
- Sub-agents complete successfully (their work is correct) but the overall result is incomplete
The key diagnostic: if sub-agents succeeded at their assigned tasks but the final output misses topics, the problem is in decomposition, not execution.
How context overflow causes information loss
Even correct decomposition can fail at scale. A coordinator correctly identifies 15 research papers, all 15 get analyzed, but when the coordinator passes all 15 analyses to the synthesis sub-agent, only 8 fit in the sub-agent’s context window. Seven analyses are silently dropped.
The coordinator is responsible for managing information flow between agents, including ensuring data fits each agent’s constraints. When results are too large, the coordinator should summarize, chunk, or extract key findings before passing them.
The iterative check
The coordinator should evaluate decomposition completeness after the first round:
- Did the subtasks cover all aspects of the original request?
- Are there domains, perspectives, or angles that were missed?
- If gaps exist, delegate targeted follow-up — don’t restart from scratch.
This evaluation-and-iteration loop catches narrow decomposition before it reaches the user. The coordinator assesses coverage, identifies gaps, and fills them incrementally.
One-liner: Sub-agents only work on what they’re assigned — narrow decomposition at the coordinator level is the root cause of missing coverage, not sub-agent execution failures. Evaluate decomposition completeness and iterate.