What is the main cost of splitting one agent into many?
Why B is the answer
More coordination, latency and failure points
Each hand-off adds calls and chances for error, so split only with a reason.
In detail
- Each hand-off adds calls and chances for error.
- Split only with a clear reason.
Use case
An illustrative scenario, not a real incident.
Splitting one agent into five adds calls, delay and more ways to fail.
See it in different ways
flowchart TD A["Supervisor"] --> B["Researcher"] A --> C["Analyst"] A --> D["Writer"] B --> A C --> A D --> A
stateDiagram-v2 [*] --> StepsKnown StepsKnown --> Workflow: yes StepsKnown --> Agent: no Agent --> Supervisor: distinct specialists Supervisor --> [*] Workflow --> [*]
mindmap
root(("Many agents"))
n1["Costs"]
n2["Coordination"]
n3["Latency"]
n4["Failure points"]Key point: Each hand-off adds cost.
Why the other options are wrong
- A. A smaller context window on every model call
- Window size is unchanged.
- C. The loss of all tool calling in the system
- Tool calling still works.
- D. A requirement to train a new model for each agent
- No new model is required.