Technical practice
How would you decide between a single-agent and multi-agent design for a public-service task?
What to coverThe role involves designing and deploying both single- and multi-agent systems. Suggested preparation: explain when a simpler design is sufficient, when specialized agents add value, and how you would manage coordination and reliability.
How would you make LLM tool use and function calling dependable?
What to coverThe posting requires hands-on experience with tool use and function calling. Suggested preparation: discuss input validation, permissions, handling failed calls and checking outcomes before an agent takes a consequential action.
How would you orchestrate a multi-step agent workflow in production?
What to coverMulti-step reasoning, orchestration and production delivery are stated requirements. Suggested preparation: walk through state management, retries, stopping conditions and recovery when an intermediate step fails.
How would you evaluate an agent whose outputs vary between runs?
What to coverThe role calls for evaluation frameworks for non-deterministic AI systems. Suggested preparation: describe representative test cases, measures of task success and unsafe behavior, and how you would use results to compare changes.
What would you monitor to diagnose unreliable agent behavior after deployment?
What to coverThe posting calls for observability tools that make production agent behavior measurable and reliable. Suggested preparation: explain what workflow and tool-call evidence you would capture, how you would investigate a failure, and how you would avoid exposing sensitive information.
How would you design guardrails for an agent supporting public-service delivery?
What to coverThe role explicitly includes production guardrails, particularly for regulated environments. Suggested preparation: distinguish checks before and after agent actions, and explain how you would test that safeguards catch harmful behavior without unnecessarily blocking useful work.
How would you set autonomy levels, human checkpoints and escalation rules?
What to coverEstablishing autonomy governance, human-in-the-loop checkpoints and escalation protocols is a stated responsibility. Suggested preparation: relate the level of human review to the risk of an action and explain what happens when the agent is uncertain or a review is not completed.
How would you choose an agent technology stack for a client use case?
What to coverThe role involves selecting and integrating agent harnesses, IDE-based agents, cloud agents and orchestration frameworks. Suggested preparation: compare options against the use case’s integration, operational and governance needs rather than assuming one framework fits every engagement.
When would retrieval-augmented generation improve an agent, and how would you assess it?
What to coverRAG and vector databases appear as preferred, not required, experience. Suggested preparation: explain when retrieval is useful, how source quality affects answers, and how you would evaluate whether retrieval improves the task.
What would a reusable reference architecture for agent delivery include?
What to coverThe role calls for reference architectures, tools and patterns that scale across public-sector engagements. Suggested preparation: separate reusable components from client-specific decisions and explain how production evaluation, observability and governance would fit into the design.
Behavioural practice
Tell me about a time you clarified an ambiguous client-facing technical problem.
What to coverThe posting calls for independent work with clients amid ambiguity. Suggested preparation: use a real situation, explain how you established requirements and aligned stakeholders, and state the resulting decision or delivery outcome.
Describe a time you communicated a technical trade-off to stakeholders.
What to coverCommunication and stakeholder management are required for this client-facing role. Suggested preparation: identify the competing priorities, your explanation and recommendation, and the decision or result.
Give an example of coaching a delivery team on an engineering practice.
What to coverCoaching teams on agentic engineering practices is a stated responsibility. Suggested preparation: describe the team’s starting point, what you changed or taught, and the observable result without claiming a specific employer process.
Tell me about a capability you made reusable beyond one delivery effort.
What to coverThe role emphasizes repeatable capabilities across engagements, and reusable tools or accelerators are preferred experience. Suggested preparation: explain the original problem, how you made the solution adaptable, and evidence that others could use it.
Describe a time you balanced delivery progress with a safety or governance concern.
What to coverThe posting emphasizes safe, reliable agents and governance in regulated environments. Suggested preparation: use a real situation to explain the risk, who you involved or escalated to, the action you took and its result.