CCAR-P Advanced Agentic Architecture Practice Question
A claims-processing agent runs for hours and must survive process restarts without losing in-flight work. You are designing durable execution around Claude's stateless Messages API. Which TWO practices are required to make the agent resumable? (Choose two.)
⚠ Common exam trap
The trap here is assuming the API keeps conversation state or that caching responses provides durability, when only externally persisted history plus in-flight checkpoints enables safe resumption.
Answer choices
Why each option matters
Answer the question above first, then reveal the full breakdown to understand why each option is right or wrong.
Correct answer & explanation
✓
Persist the full message history and pending tool_use ids to durable storage after each turn so the loop can be reconstructed.
Because the Messages API is stateless, resumability is entirely the orchestrator's responsibility. You must persist the conversation, including tool_use and tool_result pairings, and checkpoint in-flight tool dispatches so a restart does not replay side effects. Together these produce a replayable log that reconstructs the loop and reconciles uncertain operations, turning crash recovery into a deterministic continuation.
Answer analysis
Option-by-option breakdown
For each option: why learners choose it and why it is or isn't the right answer here.
- ✓
Persist the full message history and pending tool_use ids to durable storage after each turn so the loop can be reconstructed.
Why this is correct
The Messages API is stateless, so the entire conversation including assistant tool_use blocks and their matching tool_result blocks must be stored externally. On restart, replaying this history lets the loop continue exactly where it stopped. Without persisting the tool_use ids, you cannot correctly pair results and the API will reject or misinterpret the reconstructed conversation.
- ✗
Rely on the model's server-side session state to remember prior turns across restarts.
Why it's wrong here
The Claude Messages API does not maintain server-side conversation state between requests; each call is independent and the caller supplies all context. Assuming hidden session persistence leads to lost context after a restart and incorrect continuations. Durability must be implemented by your orchestration layer, not delegated to the model provider, which is why explicit persistence of history is necessary.
- ✗
Cache every model response in a CDN so identical prompts return instantly after a restart.
Why it's wrong here
CDN caching serves identical HTTP responses quickly but cannot reconstruct a partially completed agent loop or reconcile side effects. Two runs with the same prompt but different tool results would diverge, and cached completions may be stale. Caching optimizes repeat reads; it does not provide the ordered, mutable checkpoint log that resumable execution requires.
- ✓
Store a checkpoint that records which tool calls were dispatched but whose results were not yet appended.
Why this is correct
A crash between executing a side-effecting tool and appending its tool_result creates ambiguity: replaying blindly can double-charge or double-send. Recording dispatched-but-unresolved calls lets the resumed loop query or reconcile those operations before continuing. This checkpoint is what makes at-least-once execution safe by turning it into effectively-once behavior at the business level.
- ✗
Increase the context window by enabling extended thinking so the agent retains more history internally.
Why it's wrong here
Extended thinking allocates more compute to reasoning on a single request; it does not create durable storage or survive process restarts. The reasoning is not a substitute for persisted conversation state, and thinking blocks are not a resumable log. Relying on it for durability confuses per-request reasoning depth with cross-restart persistence, which are unrelated concerns.
About these practice questions
This CCAR-P question is part of Courseiva's 262-question bank — original exam-style content with full explanations and wrong-answer analysis, never real exam questions or exam dumps. Learn why practice questions differ from exam dumps →
JA
Written and reviewed by Johnson Ajibi, MSc IT Security
Senior Network & Security Engineer · founder of Courseiva
Last reviewed September 2026 · checked against the official Anthropic exam blueprint
This CCAR-P practice question is part of Courseiva's free Anthropic certification practice question bank. Courseiva provides original exam-style practice questions with explanations, topic-based practice, mock exams, readiness tracking, and study analytics to help learners prepare for the CCAR-P exam.