What this domain covers
This is the largest-weighted material you will face in this area of the Claude Certified Architect exam. Expect scenario-based questions that test whether you can apply these concepts, not just recall them.
- The agentic loop (gather context, take action, verify work)
- Workflows vs agents and the start-simple principle
- Prompt chaining, routing, parallelization, orchestrator-workers, evaluator-optimizer
- Multi-agent coordination and subagent spawning
- Session and state management, checkpointing, resumability
- Single-agent vs multi-agent trade-offs for production
- Agent guardrails: permissions, sandboxing, stopping conditions
Free practice questions: Agentic Architecture & Orchestration
8 original, exam-style questions from this domain. Reveal each answer for the full explanation.
Which type of task benefits MOST from being split across parallel subagents?
- AA tightly sequential task where each step depends on the output of the previous step
- BA task where every agent must continuously see the other agents' evolving intermediate state
- CA quick, targeted edit to a single known file where response latency matters most
- DRead-heavy research where independent sources can be explored simultaneously and each subagent returns a condensed summary
▸Show answer & explanation
Answer: D. Read-heavy research where independent sources can be explored simultaneously and each subagent returns a condensed summary
Parallel subagents shine on parallelizable, self-contained, read-heavy work: each worker explores independently in its own context and returns compressed findings. Sequential dependencies eliminate the parallelism benefit, tasks requiring continuously shared state incur heavy coordination overhead, and quick targeted edits are better done in the main conversation because subagents start fresh and add latency.
Read the docs →What happens when a PreToolUse hook script in Claude Code exits with code 2?
- AThe tool call is blocked, and the hook's stderr output is fed back to Claude as an error message
- BThe tool call proceeds, but a warning banner is shown to the user in the transcript
- CThe tool call is blocked, and the hook's stdout JSON is parsed for a permissionDecision value
- DThe hook is retried once before Claude Code falls back to the normal permission prompt
▸Show answer & explanation
Answer: A. The tool call is blocked, and the hook's stderr output is fed back to Claude as an error message
Exit code 2 from a PreToolUse hook is a blocking error: the tool call is prevented and the stderr text is returned to Claude so it understands why the action was rejected. Stdout, including any JSON, is ignored on exit code 2; JSON decisions like permissionDecision are only processed when the hook exits with code 0. There is no automatic retry behavior for hooks.
Read the docs →Which sequence correctly describes the core loop Anthropic uses to describe how agents operate?
- AClassify the input, route it to a specialized prompt, then aggregate the responses.
- BPlan the full task upfront, execute every step, then summarize and terminate.
- CGenerate a draft, have a second model evaluate it, then return the first draft unchanged.
- DGather context, take action, verify work, then repeat.
▸Show answer & explanation
Answer: D. Gather context, take action, verify work, then repeat.
Anthropic describes the agent loop as gathering context, taking action, verifying work, and repeating until the task is done. Classify-and-route describes the routing workflow, not the agent loop, and generate-then-evaluate describes evaluator-optimizer. Planning everything upfront and terminating contradicts the iterative, feedback-driven nature of agents.
Read the docs →Which two variations of the parallelization workflow does Anthropic describe in 'Building effective agents'?
- ARouting, which classifies inputs, and gating, which validates intermediate outputs between steps.
- BSharding, which splits large documents across context windows, and batching, which groups API requests to reduce cost.
- CMapping, which transforms each input item independently, and reducing, which merges partial results into one summary.
- DSectioning, which breaks a task into independent subtasks run in parallel, and voting, which runs the same task multiple times to get diverse outputs.
▸Show answer & explanation
Answer: D. Sectioning, which breaks a task into independent subtasks run in parallel, and voting, which runs the same task multiple times to get diverse outputs.
Anthropic names exactly two parallelization variants: sectioning (independent subtasks in parallel) and voting (the same task run multiple times for diverse outputs). Routing is a separate workflow pattern and gates belong to prompt chaining, not parallelization. Sharding, batching, and map-reduce are general distributed computing terms that Anthropic does not use for this pattern.
Read the docs →Which statement accurately describes how the Anthropic memory tool (memory_20250818) executes its file operations?
- AIt is a client-side tool: Claude requests operations such as view and create, and the developer's application executes them against storage the developer controls.
- BIt is a server-side tool: Anthropic hosts the /memories directory and persists the files between conversations automatically.
- CIt stores memory entries in an Anthropic-managed vector database and retrieves them by semantic search at the start of each session.
- DIt requires the code execution tool, because memory files are written inside Anthropic's sandboxed container.
▸Show answer & explanation
Answer: A. It is a client-side tool: Claude requests operations such as view and create, and the developer's application executes them against storage the developer controls.
The memory tool operates client-side: Claude only emits tool_use requests, and your application executes each command against its own storage, returning results in tool_result blocks. The /memories path is just a prefix your handler maps onto real storage such as a per-user directory or database keys. Anthropic does not host the files, there is no managed vector database behind it, and it has no dependency on the code execution tool.
Read the docs →Which of the following is a genuine advantage of a single-agent architecture over a multi-agent architecture in production?
- AIt reliably produces better results on breadth-first research tasks whose information exceeds a single context window.
- BIt eliminates non-determinism, since a single agent produces identical outputs across identical runs.
- CIt removes the need for observability tooling, because errors cannot compound within one agent.
- DFailures are easier to trace and debug because there is one context and one decision-making loop instead of coordination across multiple agents.
▸Show answer & explanation
Answer: D. Failures are easier to trace and debug because there is one context and one decision-making loop instead of coordination across multiple agents.
With one agent there is a single trajectory to inspect, so debugging and tracing are simpler and cheaper than reconstructing interactions among an orchestrator and multiple subagents. Breadth-first research exceeding one context window is exactly where multi-agent systems excel, single agents remain non-deterministic across runs, and long-running single agents are still stateful systems where errors compound, so observability is still required.
Read the docs →According to Anthropic's documentation, in which environment is it appropriate to run Claude Code in bypassPermissions mode?
- AOn any developer workstation, as long as CLAUDE.md instructs the agent to avoid destructive commands
- BOn production servers, provided an on-call engineer is monitoring the session
- CIn any git repository, because uncommitted changes can always be recovered from the working tree
- DIn an isolated environment such as a container or VM where the agent cannot damage systems that matter
▸Show answer & explanation
Answer: D. In an isolated environment such as a container or VM where the agent cannot damage systems that matter
The docs warn that bypassPermissions skips permission prompts and should only be used in isolated environments like containers or VMs where Claude Code cannot cause damage. CLAUDE.md instructions shape model behavior but are not an enforced boundary, so they do not make the mode safe on a workstation. Monitoring a production session does not contain the blast radius of a mistaken action, and git does not protect files outside the repository or actions like network calls.
Read the docs →Which statement best describes the fundamental tradeoff Anthropic identifies for agentic systems compared with a single LLM call?
- AAgentic systems reduce both latency and cost because work is split across smaller, cheaper model calls.
- BAgentic systems often trade higher latency and cost for better performance on complex tasks, so teams must evaluate when that tradeoff makes sense.
- CAgentic systems trade accuracy for speed, so they should be reserved for tasks where correctness is not critical.
- DAgentic systems remove the need for evaluation because the agent verifies its own work at every step.
▸Show answer & explanation
Answer: B. Agentic systems often trade higher latency and cost for better performance on complex tasks, so teams must evaluate when that tradeoff makes sense.
Anthropic states that agentic systems often trade latency and cost for better task performance, and developers should consider when this tradeoff makes sense. Splitting work across calls increases rather than decreases total tokens and wall-clock time, agents are used precisely because complex tasks demand correctness, and autonomous behavior increases (not removes) the need for evaluation and guardrails.
Read the docs →Drill this domain in Claude Prep
The app has dozens more questions in this domain alone, plus a Domain Drill mode that focuses your practice exactly here, and full mock exams that weight it at 27%.