Kairox Arch Prep
EXAM SCENARIO

Multi-agent research system

How the multi-agent research system scenario appears on the CCA-F exam: coordinators, subagent briefs, parallelism, citations and conflicts.

The multi-agent research system scenario puts you in charge of a coordinator agent that splits a research question across subagents, each exploring part of the problem, and then assembles a report. It draws mainly on agentic architecture and context management: how work is divided, how results flow back, and how the system stays accurate as information piles up.

The system you should picture

A user asks a broad question — the state of a market, a technical comparison, a literature review. The coordinator decomposes it into sub-questions, writes a brief for each subagent, and dispatches them. Each subagent searches and reads in its own context window, then returns findings to the coordinator, which reconciles them and writes the final answer with citations.

Decisions the exam probes

Is multi-agent worth it? Several agents cost more tokens and add coordination overhead. They earn that cost when the work is broad, parallelisable and too large for one context window. A short task with one tool does not need them.

The brief. Subagents start without the coordinator's context, so delegation has to say what success looks like: the objective, the scope boundaries, which sources to use, the format of the result and when to stop. Vague briefs produce vague findings.

What comes back. Subagents should return compact, structured findings rather than raw transcripts, so the coordinator's context does not overflow. Each claim travels with its source and the location of the supporting passage, so the final report can be verified.

Parallel or sequential. Independent sub-questions run in parallel to cut latency; steps that depend on earlier results run in order.

Conflicts and drift. When subagents disagree, both claims are kept with their sources and either checked further or reported as unresolved. Over long sessions, a compact record of established findings keeps early results from being contradicted later.

Common traps

How to practise

Write a coordinator that sends two subagents to research different halves of a question and requires each to return a list of claims with sources. Then make the two disagree on purpose and decide how your coordinator should handle it. Measure how much context the coordinator uses with raw transcripts versus structured findings — the difference is usually dramatic.

The six scenarios are listed in Anthropic's official exam guide (version 1.1), with four drawn per sitting. The emphasis described here is our own study guidance, not official exam content.

Practice questions set in this scenario

Original questions, not real exam items. All 12 questions in this scenario also appear in the practice exam.

Practice question 1

Domain 1: Agentic Architecture & Orchestration

A research system uses one coordinator and several subagents that each search a different source. Subagents return their full raw search transcripts, and the coordinator's context fills up after three subagents. Which change best addresses this?

  1. A. Cap each subagent at a single search query so its transcript is always short enough
  2. B. Increase the coordinator's max_tokens so it has more room to hold the incoming transcripts
  3. C. Run every search inside the coordinator's own conversation so all evidence stays together
  4. D. Have each subagent return a compact, structured summary of findings with their sources
Show answer and explanation

Answer: D. Context isolation is a key benefit of subagents: each explores in its own window and hands back only what the coordinator needs. Distilled, structured findings with source references keep the coordinator small while preserving traceability. Merging everything into one conversation defeats the purpose; max_tokens controls output length, not how much context is available; one query per subagent cripples research quality.

Practice question 2

Domain 1: Agentic Architecture & Orchestration

In a coordinator–subagent research pipeline, three subagents investigate independent sub-questions and none depends on another's result. Latency is the main complaint. What should the coordinator do?

  1. A. Run the subagents one after another so each can build on the previous subagent's context
  2. B. Replace the three subagents with one agent that answers every sub-question in a single turn
  3. C. Run the fastest subagent first and use its findings to brief the other two subagents
  4. D. Dispatch the independent subagents in parallel and merge their results when all return
Show answer and explanation

Answer: D. Independent work is the textbook case for parallel fan-out: total latency approaches the slowest subagent rather than the sum. Sequential execution — including briefing later subagents with earlier findings — only helps when later steps depend on earlier results, which the stem rules out. Collapsing into one agent loses context isolation and focus.

Other scenarios