Question 1 · 6.1
A retail bank’s COO asks your team for “a Claude agent that answers all customer emails”. She wants a working demo at next month’s board meeting. Nobody from the contact centre or compliance has been involved.
What should the architect do first?
A Build a demo agent on public FAQ content so the board sees early progress. B Pick the most capable model so quality is not a risk in the demo. C Run discovery with contact-centre and compliance staff; agree success criteria. D Send the COO a questionnaire and design from her answers. Question 2 · 6.1
During discovery for a logistics company, the sponsor says the new document-processing system “must be highly accurate and fast”.
Which two follow-ups turn this into usable requirements? (Select 2.)
A Agree a per-field accuracy target measured on a labelled set of real documents. B Ask whether results are needed live or can be produced in an overnight batch. C Record “high accuracy” as a non-functional requirement for the design team. D Promise 99.9% accuracy so the sponsor is confident in the project. E Defer accuracy targets until the pilot shows what the model can do. Question 3 · 6.1
In discovery for a hospital scheduling assistant, the operations director wants the assistant to rebook appointments automatically. The clinical safety lead insists a person approves any change to urgent appointments.
How should the architect handle this?
A Design full automation, since the operations director is the project sponsor. B Log it as an open decision and have its owner choose between options. C Add human approval everywhere to be safe and move on. D Ask the model to judge which appointments are urgent and route accordingly. Question 4 · 6.2
You recommend a fixed workflow rather than an autonomous agent for an insurer’s claims triage. The CIO asks why, since “agents are the future”.
Which response best communicates the decision?
A Agents are unreliable, so we avoid them in regulated industries like insurance. B It met the agreed accuracy target; an agent adds cost and latency we can’t yet justify. C We can build an agent instead if you prefer; both approaches would work fine. D Here is a detailed diagram of the workflow’s prompt chain and retry logic. Question 5 · 6.2
A hospital’s board must approve an architecture for drafting discharge letters. You have a 40-page technical design.
Which two things should the summary for the board contain? (Select 2.)
A The decision needed from the board and the deadline for it. B Only the benefits, so the board is not unduly worried about AI. C The full prompt text and retry settings for each pipeline step. D The residual risk of the chosen design and how it is controlled. E A statement that the model will always produce correct letters. Question 6 · 6.2
Eighteen months after launch, a new engineering manager at a BPO proposes replacing the complexity-based model router with the largest model everywhere, “to keep it simple”.
What would most help the team evaluate this proposal?
A Asking the original architect to explain the decision from memory. B Approving the change, since simplicity is a core design principle. C Rejecting the change because the architecture was already signed off. D The ADR recording the original context, decision and consequences. Question 7 · 6.3
An insurer’s procurement team asks your firm to sign an SLA guaranteeing 99.99% availability for a claims assistant built on the Claude API’s standard tier.
What is the best response?
A Sign it, since the Claude API is highly reliable in practice. B Sign it but exclude any outage caused by third-party services. C Refuse any SLA, since AI systems cannot offer service levels. D Propose a level based on measured SLIs, dependencies and fallbacks. Question 8 · 6.3
During a monthly review, a hospital’s patient-services manager says the appointment assistant feels slow. The team’s dashboard shows average latency of 1.2 seconds.
What should the architect do next?
A Look at the 95th- and 99th-percentile latency and when the slow requests happen. B Explain that 1.2 seconds is fast, so the perception is mistaken. C Switch to the fastest available model straight away. D Add an SLA penalty clause to show the team takes latency seriously. Question 9 · 6.3
A retailer’s product team sends a steady stream of feedback about its Claude-powered returns assistant.
Which two practices keep expectations aligned over time? (Select 2.)
A A fixed-cadence review using the same scorecard of agreed metrics. B Implementing each request as it arrives to show responsiveness. C Triaging feedback into defects, test gaps, new scope and noise. D Waiting for a formal complaint before discussing performance. E Keeping quality metrics internal so the client focuses on uptime. Question 10 · 6.4
A bank’s risk committee wants to understand a new Claude-based complaints-triage system before approving it. None of the committee members are engineers.
Which diagram is most appropriate to lead with?
A A component diagram of the triage service’s internal modules. B A system context diagram of users and external systems. C A class diagram generated automatically from the codebase. D A sequence diagram of every API call in a single triage request. Question 11 · 6.4
A year after handover, an insurer decides to replace its fixed extraction workflow with an agent. The original ADR for the workflow is still in the repository.
What should happen to that ADR?
A Delete it, so nobody is confused by an outdated decision. B Edit it in place to describe the new agent design. C Mark it superseded and link it to a new ADR for the agent. D Leave it unchanged; ADRs never need updating after acceptance. Question 12 · 6.4
You are preparing implementation guidance for a retailer’s engineers, who will build the returns assistant you designed.
Which two items give them the most actionable guidance? (Select 2.)
A An acceptance test set with a scoring method and pass mark. B A slide deck summarising the vision for the assistant. C Versioned prompt files and tool definitions in the repository. D A note to “follow prompt engineering best practice”. E A recording of the design workshop for reference. Question 13 · 6.5
Your team is two weeks from handing a claims-extraction system to an insurer’s IT department. The documentation is complete, but the insurer’s engineers have not yet made any change to the system.
What is the most important remaining handoff activity?
A Add more detail to the design document so nothing is missed. B Extend the partner’s support contract to cover the first year. C Have the insurer’s engineers ship a real change through the eval suite. D Record a video walkthrough of the architecture for later reference. Question 14 · 6.5
Anthropic emails a BPO that the Claude model behind its contact-centre assistant is deprecated. The BPO has a maintained 500-case eval suite.
Which two actions should come first? (Select 2.)
A Audit usage to find every application calling the deprecated model. B Switch all traffic to the newest model the same day. C Run the eval suite against the recommended replacement model. D Wait until the retirement date is close before acting. E Ask Anthropic to keep the model available indefinitely. Question 15 · 6.5
Three months after launch, a bank’s complaint-summary tool shows a falling “accepted without edits” rate among complaint handlers — from 88% to 74% — while latency and uptime are unchanged.
Which lifecycle response is best?
A Treat it as healthy, since the SLOs for uptime and latency are met. B Investigate the edits, update the tests, then change and re-test. C Rewrite the system prompt in production to stop the decline quickly. D Replace the workflow with an autonomous agent to handle new cases.