Paper 2 · Question 15

A: Task Statement 2.1 does teach descriptions as the first fix for misrouting, which is why this option is tempting, but that rule assumes the agent has a workable number of tools and simply cannot tell two of them apart. Here there are 22, well past the 4-5 range where selection stays reliable. Decision complexity degrades selection regardless of how well each description is written, so rewriting all 22 treats the symptom and leaves the cause.

B: Tool overload at 22 tools degrades selection reliability. Splitting into role-specific agents with 4-5 tools each reduces decision complexity and scopes each agent to its specialisation. The query agent can focus on selecting the right data source without being distracted by transformation or export tools.

C: Forced tool selection based on a pre-processing step is over-engineered and brittle. It bypasses the LLM's ability to reason about which tool to use and breaks when queries span multiple data sources.

D: A universal query tool hides the data source distinction from the agent. When the agent needs to reason about which data source contains the relevant data, it loses the ability to make that decision explicitly.