LIVE WORKSHOP · 15 MINUTES

Build an AI PM
Agent

A rapid-fire sprint to build an AI Agent Type 2 with MCP servers using Claude — from zero to a working PM agent that writes PRDs and creates Jira tickets.

📖 Learn about the 7 Types of AI Agents →

5Steps
15Minutes
1PRD
6+Tickets
Progress
0%
Step 1 of 5 ⏱ 2 minutes

Welcome — Build an AI PM Agent in 15 Minutes

This is a hands-on sprint. You'll leave with a real, reusable AI Agent Type 2 powered by MCP servers — a Claude-based PM agent that writes PRDs and creates Jira tickets through live tool integrations.

🤖 What is an AI Agent Type 2 with MCP?

This is a tool-using AI agent — Claude configured with a PM system prompt that can call live external tools via MCP (Model Context Protocol) servers. It doesn't just chat — it takes real actions: writing PRDs, creating Jira tickets, reading your calendar, and pushing to Slack. Learn about all 7 Types of AI Agents →

What you'll build in 20 minutes
⚙️
PM System PromptConfigure Claude to think and act like a Senior PM in any session
📋
Full PRDProblem statement, user stories, and quantifiable success metrics
🎫
Jira BacklogEpic + Stories + Tasks with full acceptance criteria — ready to assign
⚠️
Risk AnalysisSeverity-ranked risks with concrete mitigation strategies
Before you start — tick these off
  • claude.ai is open in another tab, ready to receive prompts
  • You have a feature idea in mind — even rough and half-baked is perfect
  • (Optional) Jira project key handy if you want live ticket creation at the end

Hit "Begin Sprint" to start the 15-minute countdown. Each step shows how many minutes to spend. Move fast — this is designed to be tight but totally doable.

Step 2 of 5 ⏱ 5 minutes

Configure Your PM Agent

This system prompt is the DNA of your agent. Paste it into Claude as your first message — it transforms Claude from a generalist into a disciplined Senior PM for the rest of the session.

3 steps to get your agent running
  • 1
    Switch to your Claude.ai tab and open a new conversation.
  • 2
    Click "Copy Prompt" on the block below and paste it as your first message to Claude.
  • 3
    Wait for Claude to reply with: "I'm your Senior PM Agent. What feature are we building today?" — then come back here.
Copy this into Claude now
SYSTEM_PROMPT.txt — Paste as first message
You are a Senior Product Manager AI Agent with 10+ years of experience at top tech companies. Your responsibilities in this session: 1. CLARIFY — Ask targeted clarifying questions before proceeding. Never skip. Ask 4–6 questions covering: target users, core problem, scope, success metrics, and constraints. 2. GENERATE PRD — After clarification, produce a structured PRD: - Problem Statement (specific, user-focused, 2–3 sentences) - Target Users (primary and secondary personas with context) - User Stories ("As a [user], I want [action] so that [outcome]" — minimum 5) - Success Metrics (measurable, 90-day targets) - Out of Scope (explicit v1 exclusion list — no vague "future enhancements") 3. CREATE JIRA TICKETS — Convert PRD into: - 1 Epic (full context, goals, success metrics) - 4–6 Stories (user-facing, acceptance criteria as checkboxes, 5–8 per story) - 1–2 Tasks (technical spikes, definition of done) - Story points: Fibonacci scale (1, 2, 3, 5, 8 — nothing over 8) 4. FLAG RISKS — 3–5 risks: name, severity (High/Med/Low), description, mitigation, owner. 5. ITERATE — Ask if anything needs refinement. Rules: Direct, structured, no filler. Flag assumptions with "Assumption:" prefix. Begin by saying: "I'm your Senior PM Agent. What feature are we building today?"

You'll know it worked when Claude replies: "I'm your Senior PM Agent. What feature are we building today?" — Switch back here and click Next.

Step 3 of 5 ⏱ 5 minutes

Describe Your Feature & Answer Questions

Tell the agent what you want to build using the template below. It will ask 4–6 targeted questions — answer them honestly and specifically. This dialogue is what makes the PRD great.

The right level of detail
❌ Too vague

"I want to improve our scheduling feature."

✅ Just right

"I want a Chrome sidebar for Zara AI so EAs can schedule meetings without leaving Gmail."

Fill in the brackets, then paste into Claude
FEATURE_INPUT.txt — Customize and send
I want to build: [DESCRIBE YOUR FEATURE IN 1–3 SENTENCES] Context: - Our product: [BRIEFLY DESCRIBE YOUR PRODUCT] - The pain point: [WHAT PROBLEM DOES THIS SOLVE?] - Who it's for: [PRIMARY USER — be specific, e.g. "EAs at B2B SaaS companies managing 3+ execs"] That's all I have — help me shape this into a proper product spec.
Tips for answering the clarifying questions
Be specific about users
"EAs managing 3+ executives at Series B startups" beats "busy professionals."
Scope down hard on v1
When asked about scope, cut aggressively. A thin v1 ships. A bloated v1 doesn't.
"I don't know" is fine
The agent will flag it as an assumption. You can override it with one message later.
Name your user
"Sarah, EA at a 200-person startup" gives the agent far more to work with.
🎯

After your answers, Claude will say something like "I have enough context to write the PRD." That's your cue to come back here and hit Next.

Step 4 of 5 ⏱ 4 minutes

Generate the PRD

One prompt triggers the full Product Requirements Document. Review it quickly, use the refine prompt if anything's off, then move to tickets.

Send this to Claude now
PRD_GENERATION.txt
Generate the full PRD based on everything we've discussed. Sections: 1. Problem Statement (specific, user-focused, 2–3 sentences) 2. Target Users (primary and secondary personas with context) 3. User Stories (minimum 5, format: "As a [persona], I want [action] so that [outcome]") 4. Success Metrics (measurable, 90-day targets — e.g. "DAU/MAU ≥ 40%", "500 installs") 5. Out of Scope for v1 (explicit list — no vague "future phases") Every line should be actionable by a developer or designer. No filler.
Quick-fix prompt if something needs adjusting
REFINE_PRD.txt — Optional
Refine the PRD: - [e.g. "Make the problem statement specific to EAs, not executives"] - [e.g. "Add a user story covering the onboarding flow"] - [e.g. "Reduce the 90-day install target to 200"] Keep everything else the same.
PRD quality checklist
  • Problem Statement names a specific user and a specific pain
  • User Stories use full "As a / I want / So that" format
  • Success Metrics are numbers — not vague phrases like "improve engagement"
  • Out of Scope is an actual list — not "future enhancements"
Step 5 of 5 ⏱ 4 minutes

Convert to Jira Tickets + Risks

The final move — turn your PRD into a complete sprint backlog with a risk analysis, all in one shot. Optionally push directly to Jira via the API.

Ticket + risk generation prompt
JIRA_TICKETS.txt
Now convert the PRD into Jira tickets. Create: - 1 Epic: full description, goals, success metrics, out of scope - 4–6 Stories (one per user story): summary, user story sentence, background (2–3 sentences), acceptance criteria (5–8 checkboxes), story points - 1–2 Tasks for technical spikes: what to validate, why it matters, definition of done (3–5 items), story points Story points use Fibonacci: 1, 2, 3, 5, 8 — split anything over 8. After the tickets, add a risk analysis: - 3–5 risks sorted by severity (High first) - Each: name, description, mitigation action, owner (Eng/Product/Design/Legal) Format each ticket and risk clearly with headers.
🔗

If you have Jira connected to Claude: After reviewing the tickets, send the prompt below to push them live to your Jira project.

JIRA_PUSH.txt — Optional live creation
The tickets look good. Create them in Jira now. Project key: [YOUR_PROJECT_KEY] Jira site: [yourcompany.atlassian.net] Create the Epic first, then all Stories and Tasks under it. Confirm each ticket number as you create them.
Final quality checklist
  • Every Story has a user-facing outcome (not just technical tasks)
  • Acceptance criteria are specific and testable — not subjective
  • No single story exceeds 8 story points
  • At least one technical spike exists for your highest architectural risk
  • High-severity risks have a named owner and concrete action
🏆

You just did in 20 minutes what takes most teams 1–2 weeks: feature scoping, full PRD, complete ticket backlog, and risk analysis. Hit Complete to see your reusable toolkit.

🎉

Workshop Complete!

You built and used a working AI PM Agent. Here's what you created in 20 minutes:

5Steps Done
1Full PRD
~8Jira Tickets
20Minutes
Your reusable AI PM Agent toolkit
⚙️
System PromptPaste into any new Claude conversation to instantly activate your PM Agent
📋
PRD FrameworkThe same 5-section structure works for any feature in any sprint
🎫
Ticket GeneratorOne prompt → full sprint backlog with acceptance criteria every time
⚠️
Risk AnalyzerRun the risk prompt at the start of every sprint to stay ahead of blockers
What to do next
  • 🚀
    Run it on a real backlog item this week — take something your team is debating and go through the full workflow.
  • 🤝
    Share this workshop with engineers and designers so your whole team speaks the same PRD language.
  • ⚙️
    Customize the system prompt to match your company's PRD format, ticket naming conventions, and Jira project structure.
💙

Built by TeamCal AI — AI-powered team scheduling software trusted in 90+ countries. Meet Zara, your AI executive scheduling partner at teamcal.ai/getzaraai →