20 Questions — Domain 6 : Stakeholder Communication & Lifecycle Management Quiz

Expand any question to reveal the correct answer and explanation.

1. 1 A senior executive asks for an 'AI agent to automate all legal reviews' without specific metrics. According to the Professional Lifecycle guidelines, what is the most appropriate first output to deliver?

   Consider the sequence of lifecycle phases and what defines a successful start to the engagement.

   A business problem statement with measurable success criteria and a current-process baseline.

   The discovery phase must first produce a structured analysis of the business problem, success targets, and a baseline to measure eventual improvements.

   • ✗ A functional prototype using a multi-agent framework to demonstrate feasibility.

     Jumping to a proof of concept before establishing success criteria and constraints is identified as a 'skip-discovery' trap in professional architecture.

   • ✗ An Architecture Decision Record (ADR) justifying the choice of the Claude Opus model.

     ADRs are intended to capture technical decisions made during the design phase, not to replace the requirements gathering required in discovery.

   • ✗ A detailed project plan and budget estimate for the development of a RAG pipeline.

     Proposing a specific technology solution like RAG is premature until discovery identifies the actual data access levels and technical constraints.

2. 2 When defining Service Level Agreements (SLAs) for a Claude-powered system, why should an architect favor probabilistic targets over promises of deterministic perfection?

   Think about the underlying nature of large language models and how they differ from traditional scripted code.

   Because cognitive systems produce a natural error rate that requires an error budget and remediation paths.

   LLM systems are fundamentally probabilistic; committing to a measured error rate with remediation is the standard for professional architect-level SLAs.

   • ✗ To hide the inherent instability of foundational models from non-technical stakeholders.

     Transparency is an ethical AI pillar, and framing targets should be based on technical reality rather than an attempt to obscure model behavior.

   • ✗ To align with the specific latency requirements of the Claude Haiku fast-tier model.

     Probabilistic targets refer to the likelihood of correct reasoning or output quality across any model tier, not specifically to latency metrics.

   • ✗ Because deterministic compliance is only possible through prompt engineering which is not yet verified.

     Prompt instructions are actually considered probabilistic, while programmatic layers like permission scoping provide the deterministic controls.

3. 3 Which of the following describes the correct order of the solution lifecycle phases for a production-grade Claude system?

   The professional architect must lead the system from initial ambiguity to long-term operational health.

   Discovery, Design, Handoff, Monitoring, Iteration.

   This sequence ensures requirements are gathered first, followed by technical design, transition to operations, and continuous performance feedback.

   • ✗ Design, Discovery, Implementation, Monitoring, Handoff.

     Discovery must precede design to inform architectural choices, and handoff typically occurs after design is finalized for implementation.

   • ✗ Discovery, Prototyping, Monitoring, Handoff, Iteration.

     This sequence omits the formal design and documentation phase which is critical for senior lifecycle ownership.

   • ✗ Prototyping, Design, Handoff, SLA Alignment, Iteration.

     SLA alignment should happen early in discovery and design, not as a late-stage standalone phase after handoff.

4. 4 During a stakeholder review, the Legal team expresses concern about data privacy. What discovery output is specifically designed to address this concern before architecture begins?

   Focus on the artifact that captures non-negotiable business or regulatory rules.

   A constraint inventory including compliance regimes and data residency requirements.

   Discovery must inventory business and legal constraints, such as GDPR or HIPAA, to ensure the resulting architecture is compliant by design.

   • ✗ A stratified random sampling report of historical model outputs.

     Sampling is used for evaluation and validation of an existing system, rather than identifying requirements during the discovery phase.

   • ✗ An implementation runbook detailing how to redact PII in logs.

     Runbooks are handoff documentation for operations teams, whereas discovery focuses on identifying the need for such controls.

   • ✗ A list of few-shot examples that demonstrate the model's refusal behavior.

     Prompt-level behavior is a design tactic, not a discovery output that defines the high-level legal and technical constraints.

5. 5 An architect is drafting an Architecture Decision Record (ADR) for a multi-agent system. What is the primary purpose of including 'Consequences' in this document?

   Consider what a future team needs to know about the side effects of a major design choice.

   To capture the trade-offs and secondary impacts resulting from the chosen decision.

   Documenting consequences ensures the team understands what is accepted (e.g., higher latency) in exchange for the benefits of the chosen architecture.

   • ✗ To list the specific developers responsible for any future system bugs.

     ADRs are technical reasoning tools, not mechanisms for assigning personal blame for implementation errors.

   • ✗ To provide a step-by-step tutorial for stakeholders to reproduce the design.

     Reproduction and implementation details belong in technical guides or runbooks rather than the 'Consequences' section of an ADR.

   • ✗ To define the financial penalties for failing to meet probabilistic SLA targets.

     SLA penalties are business contract terms, whereas ADR consequences refer to technical and operational effects of architectural choices.

6. 6 Which communication strategy best helps an architect justify the higher token cost of the Claude Opus tier to a non-technical budget owner?

   Think about how to link a model's superior reasoning to the company's bottom-line goals.

   Explaining how the premium tier increases transaction resolution rates and offsets its cost through operational savings.

   Translating technical performance into business value pillars like resolution rates or efficiency is the professional approach to stakeholder communication.

   • ✗ Sharing raw benchmark scores comparing Opus to Sonnet on logic-heavy tasks.

     Raw benchmark data is considered 'engineer language' and often fails to convey the actual business value of the decision.

   • ✗ Citing the exact number of parameters and tokens processed in the system prompt.

     Technical metrics like parameter counts are too granular for high-level business alignment discussions regarding ROI.

   • ✗ Suggesting that the team will switch to Claude Haiku later if the costs become too high.

     This avoids the trade-off discussion entirely rather than justifying the current architectural need for a capability-tier model.

7. 7 The 'Handoff' phase of the lifecycle requires providing documentation for the operating team. Which artifact is most critical for ensuring long-term system stability in production?

   Focus on what an on-call engineer needs to keep a system running correctly.

   Operational runbooks, evaluation baselines, and defined escalation paths.

   Runbooks and escalation paths are essential for the maintenance team to respond to incidents and verify performance against the original architecture.

   • ✗ A complete inventory of all few-shot examples used in every prompt template.

     While useful, a prompt inventory is secondary to the operational procedures required to handle system failures and maintenance.

   • ✗ The source code for the MCP server used in development.

     Code is the implementation itself; the 'handoff' artifact focuses on how to govern and maintain that implementation in a production environment.

   • ✗ A record of the initial discovery workshops with business stakeholders.

     Discovery notes provide context for the design but do not equip the operations team with the tools to manage the live system.

8. 8 In the context of stakeholder communication, how should an architect handle a request for a 100% accurate AI-driven financial audit system?

   Reflect on the professional pattern for translating probabilistic behavior into business terms.

   Set expectations around a measured accuracy floor and define human-in-the-loop escalation paths for the remaining risk.

   Architects must frame LLM behavior in terms of measured rates on an evaluation set combined with defined paths for human intervention.

   • ✗ Commit to the target by implementing a three-pass review architecture using Claude Opus.

     Committing to perfect accuracy for a probabilistic system is an 'architect error' that sets unrealistic expectations.

   • ✗ Reject the project immediately as LLMs are incapable of performing financial audits.

     While perfection isn't possible, an architect's role is to design a system that manages the error rate to meet business needs, not to reflexively reject difficult tasks.

   • ✗ Promise 100% accuracy but include a disclaimer that the system prompt might occasionally fail.

     This approach is contradictory and fails to provide stakeholders with a clear, measurable plan for managing the system's inherent probabilistic nature.

9. 9 Why is it critical for an architect to conduct 'Iteration' as a formal phase of the solution lifecycle rather than just a development task?

   Think about the role of the feedback loop in maintaining long-term architectural quality.

   To feed monitoring data and stakeholder feedback back into the design for continuous improvement.

   Iteration closes the loop by using production performance data to refine the architecture, ensuring it remains aligned with business needs over time.

   • ✗ To ensure that stakeholders are billed for additional development cycles beyond the initial launch.

     The lifecycle phase is about system health and performance alignment, not purely about billing or commercial arrangements.

   • ✗ To rewrite the entire system prompt from scratch every month to avoid model drift.

     Constant, unprincipled prompt rewriting is inefficient; iteration should be evidence-based and driven by specific monitoring signals.

   • ✗ Because non-proctored certification renewals require proof of an iterative development lifecycle.

     While certifications have renewal assessments, the architectural phase of iteration is meant to benefit the production system, not just the architect's credentials.

10. 10 During the discovery phase, you are asked to identify 'Stakeholders'. Who must be included to ensure the architecture survives compliance and security reviews?

    Consider whose approval is necessary to keep an enterprise AI deal 'alive' and compliant.

    A cross-functional group including legal, compliance, operations, and business owners.

    Professional-tier architects must lead decisions across security, legal, and executive considerations, requiring input from all these groups during discovery.

    • ✗ Only the developers who will be building the RAG pipeline.

      Focusing only on implementation ignores the senior-level concerns of governance, risk, and cross-functional ownership.

    • ✗ The end-users who will interact with the chat interface daily.

      While users provide feedback on utility, they are not typically responsible for compliance, safety controls, or business risk alignment.

    • ✗ Anthropic's Partner Academy support team.

      External support teams are resources for learning, but internal stakeholders must own the specific risk and governance of their deployment.

11. 11 What is the primary architectural purpose of a 'Success Criteria' document produced during discovery?

    Think about how an architect proves that their design actually solved the original business challenge.

    To establish the specific technical benchmarks that must be met to satisfy the business problem.

    Success criteria convert ambiguous business challenges into measurable outcomes that guide the design and later validate the system's success.

    • ✗ To provide a binding legal contract between the architect and the cloud provider.

      Contracts with providers are separate legal entities; success criteria are internal metrics to evaluate the solution's business performance.

    • ✗ To list the Claude model IDs that the team is permitted to use during development.

      Model selection is a design phase task, whereas success criteria define what the system must achieve regardless of the specific technology used.

    • ✗ To document the failure modes of the Claude Agent SDK for legal indemnity.

      Indemnity is a legal matter; success criteria are used to measure and optimize for productivity and efficiency targets.

12. 12 When presenting an architecture that uses an 'Agentic' pattern over a 'Workflow' pattern, how should the architect frame the trade-off to stakeholders?

    Consider the key performance indicators and risks associated with model-driven vs. programmatic control flows.

    Acknowledge the higher cost and latency variance in exchange for the ability to handle open-ended discovery tasks.

    Architects must be honest about the cost-performance trade-offs of autonomous agents compared to deterministic, fixed-sequence workflows.

    • ✗ Explain that the Agentic pattern is newer and therefore superior for all production workloads.

      Choosing a pattern based on novelty is a defect; extra complexity must be justified by the problem requirements.

    • ✗ Guarantee that the model's reasoning loop will always terminate within a fixed cost budget.

      Agentic loops have inherent variance in path and token consumption; promising a fixed cost is unrealistic and misleads stakeholders.

    • ✗ Argue that Workflow patterns are only for entry-level developers and do not scale to enterprise needs.

      Workflow patterns are often preferred in enterprises for their auditability and predictability, and are not restricted by developer experience level.

13. 13 Which component of an Architecture Decision Record (ADR) is specifically intended to help future architects understand why certain alternatives were rejected?

    Think about the parts of a record that justify 'the path not taken'.

    The Options Considered section.

    ADRs capture the reasoning behind a choice by documenting the context and the specific alternatives that were evaluated and ultimately dismissed.

    • ✗ The System Prompt Inventory.

      Prompt inventories list the final artifacts but do not record the reasoning behind why other prompt versions or patterns were not used.

    • ✗ The SLA Probabilistic Target.

      SLAs define performance expectations for the chosen solution, not the history of the technical decision-making process.

    • ✗ The Metadata Extraction Schema.

      Schemas are implementation details of the final design, not part of the high-level decision record regarding architectural trade-offs.

14. 14 An architect discovers during the monitoring phase that a system's latency has spiked. What is the most professional first step in communicating this to stakeholders?

    Consider the relationship between monitoring, SLAs, and evidence-based decision making.

    Present the latency data alongside the impact on current SLAs and propose evidence-based optimization levers.

    Providing quantified data and clear trade-off choices allows stakeholders to make an informed decision on whether to accept the latency or invest in optimization.

    • ✗ Immediately downsize all models to Claude Haiku to restore speed without consulting the budget owner.

      Arbitrarily changing tiers risks quality regressions and should be a reasoned decision involving stakeholder alignment on the quality-latency trade-off.

    • ✗ Blame the Anthropic API for the delay and wait for them to resolve the infrastructure issue.

      Architects are responsible for their system's performance; they must investigate local causes like context bloat or retrieval inefficiencies before assigning external blame.

    • ✗ Hide the latency spike in aggregate average metrics to avoid alarming executive leadership.

      Masking per-slice failures behind aggregate averages is a deceptive practice that undermines the transparency required in professional architecture.

15. 15 Why is 'Current-process baseline' gathering a mandatory part of the professional discovery phase?

    Think about how you prove a project was successful to an executive interested in 'transformation' or 'cost reduction'.

    Because improvement claims and productivity gains cannot be quantified without a 'before' metric.

    Stakeholders require evidence of transformation; baselines provide the data necessary to calculate ROI and verify that success criteria were met.

    • ✗ To verify that the existing human workers are performing at a lower accuracy than the proposed AI.

      Baselines are not about competing with humans, but about establishing a starting point for measuring the business value of the new system.

    • ✗ To allow the architect to copy existing human prompts into the model's system instructions.

      Human processes and AI instructions differ fundamentally; a baseline measures performance, it doesn't provide a direct architectural blueprint.

    • ✗ It is a requirement for SOC 2 and FedRAMP compliance for all AI systems.

      While compliance is important, the baseline's primary purpose in the lifecycle is for requirement gathering and value alignment.

16. 16 An architect is tasked with ensuring 'Ethical AI' considerations are met for a new recruitment tool. Which lifecycle phase is best suited for designing demographic slice evaluations?

    Consider when it is most cost-effective and responsible to define what 'fairness' means for a system.

    Discovery, to identify the bias risks and define fairness metrics.

    Identifying risks and defining measurable outcomes like fairness must happen during discovery to ensure they are integrated into the system design.

    • ✗ Iteration, to fix any biased outcomes after they have been reported by users.

      Addressing ethics only after launch is reactive; professional standards require proactive risk identification and mitigation during early phases.

    • ✗ Handoff, to inform the operations team that bias might exist in the model.

      Simply noting the risk during handoff is insufficient; the architect must design the system to actively monitor and mitigate bias.

    • ✗ Monitoring, because bias can only be detected with live production traffic.

      While monitoring is important for drift, ethical frameworks should be established early and tested against evaluation datasets before launch.

17. 17 In Domain 6, what does 'Expectation Management' primarily entail when discussing AI accuracy with stakeholders?

    Reflect on how to communicate the difference between scripted software and cognitive systems.

    Educating stakeholders that AI models are probabilistic and will occasionally produce errors even when perfectly configured.

    Professional architects must prevent 'perfection' traps by setting expectations for a measurable error rate and clear escalation paths.

    • ✗ Persuading stakeholders that accuracy is less important than system speed.

      The architect must align with the stakeholders' stated priorities (e.g., accuracy vs. speed), not unilaterally decide which is more important.

    • ✗ Training stakeholders on how to write better prompts themselves to improve the system.

      Expectation management is about the high-level capabilities and risks of the system, not offloading the technical architect's job to the stakeholders.

    • ✗ Promising that future Claude model updates will eventually eliminate all hallucinations.

      Committing to future, unverified vendor capabilities is an unreliable strategy that fails to address the risks of the current deployment.

18. 18 When documenting an architecture for an implementation team, what is the significance of 'Integration Contracts' in the handoff phase?

    Think about what is needed to ensure a complex system's parts fit together during implementation.

    They define the structured formats (e.g., JSON schemas) and protocols that different system components must use to communicate.

    Clear documentation of schemas and protocols ensures the implementation team can build components that work together reliably as designed.

    • ✗ They are the legally binding agreements signed between the company and Anthropic.

      Integration contracts in this context refer to the technical specifications for how systems communicate, not commercial legal agreements.

    • ✗ They specify the number of API tokens the developers are allowed to spend during their work hours.

      Token budgets are part of cost management, whereas integration contracts focus on the technical interface and data exchange between systems.

    • ✗ They list the employment contracts of the subagent coordination team.

      Architectural documentation focuses on system design and operation, not human resource management.

19. 19 A stakeholder feedback loop at a 'Phase Boundary' (e.g., between Design and Handoff) is intended to:

    Think about the cost of changing an architecture once implementation has fully begun.

    Identify required course corrections while they are still relatively inexpensive to implement.

    Checking alignment at boundaries prevents teams from building complex systems based on misunderstood requirements or outdated constraints.

    • ✗ Allow developers to start coding the solution without waiting for final approval.

      Feedback loops are meant to verify alignment and catch errors early, not to bypass formal phase requirements.

    • ✗ Provide an opportunity for the architect to renew their proctored certification exam.

      Certification is an individual professional milestone, whereas the feedback loop is a project governance activity.

    • ✗ Ensure that the monitoring data from the previous iteration has been deleted.

      Monitoring data should be preserved and used to inform future iterations, not deleted at phase boundaries.

20. 20 Which discovery output provides a visual map of where an agent is permitted to retrieve or write information?

    Focus on the artifact that defines the system's informational reach and boundaries.

    A data inventory and access map.

    A key discovery output is identifying what data exists, where it lives, and the specific permissions or boundaries the agent must respect.

    • ✗ The ADR consequences list.

      Consequences describe side effects of choices, they do not act as an authorization or data access map.

    • ✗ The prompt caching prefix strategy.

      Caching is a technical optimization lever used during design, not a discovery artifact defining organizational data access.

    • ✗ The Human-in-the-Loop placement diagram.

      HITL diagrams map human interaction points, whereas a data access map defines the system's underlying data boundaries.