Sierra AI·Product Manager·Onsite - System Design / Architecture·Senior

Senior

Jun 2026

Summary

Sierra AI PM interview, one question deep into agentic AI design for a real-world customer support use case. The prompt was open-ended enough that I spent the first few minutes just trying to figure out where to even start.

Questions Asked (1)

Design an agentic AI system to power customer support for Spotify.

System DesignProduct Sense & IdeationTechnical Trade-offs

Author's notes

I started with user journeys which felt right but I probably spent too long there before getting into the actual agent architecture.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by clarifying the scope and success metrics for Spotify's customer support, then design an agentic AI system that handles common issues autonomously while seamlessly escalating complex cases to human agents. Emphasize how the system leverages Spotify's unique data (listening history, subscription details) to provide personalized, proactive support, and discuss trade-offs between automation and human touch.

Pro tip: Frame your answer around Sierra AI's core value proposition: AI agents that resolve issues end-to-end, not just deflect. Show you understand that the best agentic systems are built on a deep integration with backend systems and a robust escalation policy.

1. Clarify Goals and Scope

Ask clarifying questions to define the support domains (e.g., billing, playback, account) and success metrics (e.g., resolution rate, CSAT, cost per contact). Establish constraints like latency, compliance, and integration with existing systems.

2. Map User Intents and Journeys

Identify the top customer support intents (e.g., payment issues, playback errors, account recovery) and map ideal resolution paths. Prioritize intents by frequency and complexity to determine automation potential.

3. Design Agentic Architecture

Outline the AI agent components: natural language understanding, dialogue management, tool use (APIs for account lookup, payment processing), and memory for context. Include a policy engine for decision-making and escalation triggers.

4. Define Escalation and Human-in-the-Loop

Specify when and how the AI escalates to human agents, ensuring a smooth handoff with full context. Design feedback loops where human resolutions train the AI to improve over time.

5. Measure, Iterate, and Scale

Propose metrics to evaluate performance (e.g., containment rate, resolution time, CSAT) and a roadmap for iterating from pilot to full deployment. Discuss how to scale across languages and markets.

Key Points to Mention

Integration with Spotify's backend systems (e.g., account management, payment gateways) to enable end-to-end resolution.
Use of personalization: leveraging listening history and subscription tier to tailor responses and proactive outreach.
Handling of sensitive data and compliance (GDPR, CCPA) with secure data handling and user authentication.
Multi-turn conversation capabilities and context retention for complex issues.
Clear escalation criteria and seamless handoff to human agents to maintain trust.
Continuous learning from human feedback and unresolved cases to improve the AI agent.

AI-generated suggestions, not part of the candidate's original notes. May be inaccurate — verify before relying on them.