Claude Certified Architect – Foundations Practice Question 62: Parallel vs Sequential Subagents

Claude Certified Architect, Foundations · Test 7 · question 2 of 10

Agent Architecture & Orchestrationmedium

An orchestrator agent prepares a vendor risk report. It needs a financial-health analysis, a security-posture review and a legal-terms review, and each reads different source documents and shares no state with the others. A final executive summary then needs all three results. Latency matters, and the team wants to avoid wasted spend. Which orchestration is best?

Select one answer
Show answer and explanation

Answer:

  • C. Run the three reviews as parallel subagents, then write the summary after all return

C is correct. Subagents can run concurrently, so independent subtasks finish in the time of the slowest one rather than the sum of all of them. The summary depends on all three outputs, so it must start only after they return. The Agent SDK subagents docs describe this parallelization benefit, and the prompting guide's sample instructions are consistent with it: use subagents for independent workstreams and keep sequential or state-dependent work in order.

A is wrong because it gives up the parallel speedup, and the reviews do not need shared context. B is wrong because the summary cannot be produced from placeholders; starting a dependent step early wastes tokens and yields a bad result. D is wrong because the reviews are independent, so sequencing them adds latency with no benefit.

Reference: Subagents in the SDK

Practice Test 7 ›

Study guides for Claude Certified Architect, Foundations