Office Agent PM
Take-Home Assignment

1. User scenarios & value judgment: Who is the core user of this Agent? How do different roles (IC, manager, executive, support, sales) differ in their email handling behavior? Who benefits most? Which user segment should V1 target, and why?
2. MVP trade-offs: Your team has 2 engineers and 2 months. Define what your MVP includes, what it excludes, and why:
   • (A) Priority triage — Scan the inbox and classify emails into "Act now / Later / Ignore"
   • (B) Smart reply drafts — Auto-generate reply drafts for emails that need a response
   • (C) Action item extraction — Extract action items from emails and create to-dos or calendar events
   • (D) Thread summary — Summarize long email threads so the user doesn't have to read the entire chain
   • (E) Follow-up reminders — Track sent emails that haven't received a reply and remind the user to follow up
3. Risk judgment: What happens when the Email Agent gets it wrong? Which error is more dangerous — marking an important email as "Ignore" vs. marking a routine email as "Act now"? How would you reduce these risks through product design?
4. Kill criteria: Under what conditions would you recommend not building this product, or delaying it?

For your chosen MVP scope, design the complete Agent workflow:

1. Decision logic: How does the Agent determine an email's priority and handling action? What signals does it use (sender, content, context, user history)? Which operations can be automated, which require user confirmation? What are your principles?
2. Failure modes: List at least three critical failure modes and their product-level mitigations — note that email failures are often social (wrong tone, missing a key recipient, replying at the wrong time) rather than purely technical.
3. Evaluation framework: How do you define "the Agent got it right"? What do you measure, how do you measure it, and what metrics determine whether this is ready to ship? How do you track changes in user trust?

Design a Prompt for one core capability of the Email Agent (e.g., priority classification, reply draft generation, action item extraction — pick the one you consider most critical). Include:

1. The Prompt itself
2. Expected failure scenarios — what inputs will cause this Prompt to break?
3. The iteration path from prototype to production

Using any tool of your choice (Coze, Dify, GPTs, Cursor, Python, Replit, etc.), build a working prototype of at least one core component of your Agent workflow.

Requirements:

• Choose the component you consider most critical or highest-risk — the choice itself is a product judgment
• The prototype must be runnable — we will test it; screenshots or document descriptions alone are not accepted