2.1: Write a PRD

• Time to Complete: 45-60 minutes
• Prerequisites: Module 1.4 (Agents), Module 1.5 (Custom Sub-Agents)

Start this module in Claude Code: Run /start-2-1 to kick off the interactive experience.

Overview

Module 2.1 teaches you to write better PRDs faster by using AI as a thinking partner. You’ll incorporate templates, company context, and user research via @-mentions, generate multiple strategic approaches in parallel, and get multi-perspective feedback before sharing with your team.

Key takeaway: AI is most valuable when it helps you think better, not when it does all the thinking for you. The best PRDs combine human judgment with AI’s ability to process full context, generate options, and provide diverse perspectives.

Traditional vs AI Partnership Approach

The difference: You’re never starting from scratch, and you’ve addressed potential issues before your first review.

The Four Core Techniques

1. Full Context via @-Mentions

Provide Claude with all relevant information at once.

What to @-mention:

• Company context - Product strategy, customer insights, business metrics
• User research - Interview transcripts, survey results, pain points
• Templates - Your preferred PRD structure
• Methods - Frameworks like Socratic questioning
• Related docs - Previous PRDs, competitive analysis, roadmap

Example:

Help me write a PRD for [feature] using:
@company-context.md - product and customer context
@user-research/pain-points.md - insights from 20 interviews
@prd-template.md - our standard PRD structure
@methods/socratic-questioning.md - framework for clear thinking

2. Socratic Questioning for Clarity

Use structured questioning to sharpen fuzzy thinking before writing.

Key question categories:

Pro tip: Pick 3-5 most relevant questions. Quality over quantity.

3. Generate Multiple Strategic Approaches

Have Claude generate 3 different strategic approaches to the same feature. Compare, choose the best, or combine elements.

Example: AI voice chat + to-do list feature

Using agents for parallel generation:

Spin up 3 agents to generate 3 PRD drafts in parallel:
- Agent 1: Chat-first approach
- Agent 2: List-first approach
- Agent 3: Balanced approach

This demonstrates the power of parallel processing for PM work.

4. Multi-Perspective Feedback via Sub-Agents

Get diverse feedback before sharing with your actual team.

Three key perspectives:

Example:

Get reviews from all three sub-agents and consolidate feedback:
- @.claude/agents/engineer.md - technical review
- @.claude/agents/executive.md - business review
- @.claude/agents/user-researcher.md - UX review

Real-World Example

Context setup:

I'm writing a PRD for smart meeting summaries. Help me think through this using:
@company-context.md
@user-research/calendar-pain-points.md
@prd-template.md
@methods/socratic-questioning.md

Clarify thinking:

Ask me 5 key questions about this feature using the Socratic framework:
- Problem clarity
- Why this solution
- Success metrics
- Technical constraints
- Strategic timing

Generate approaches:

Generate 3 strategic approaches:
1. Calendar-native (summaries appear in calendar app)
2. Email-first (summaries delivered via email)
3. Standalone app (dedicated meeting hub)

For each: problem framing, solution, success metrics

Get feedback:

👨‍💻 Engineer, review @smart-summaries-prd.md for technical feasibility.
Then 💼 Executive, review for business value and strategic fit.
Then 👤 User Researcher, review for user needs alignment.

Best Practices

Do:

• Start with full context (@-mention company docs, research, templates)
• Use Socratic questions to clarify fuzzy thinking
• Generate 2-3 strategic approaches, don’t go with first idea
• Get sub-agent feedback before sharing with real team
• Drive the process yourself - AI provides options, you decide

Don’t:

• Let AI write for you (you’re partnering, not outsourcing)
• Skip the thinking (questions exist to sharpen YOUR judgment)
• Accept first draft (always generate alternatives to compare)
• Ignore sub-agent feedback (often catches important issues)
• Work in isolation (use full context)

Pro Tips:

Troubleshooting

AI-generated PRDs feel generic

Cause: Insufficient context or too much AI autonomy

Fix:

• Provide more specific company context (@-mention detailed docs)
• Use Socratic questioning to inject YOUR thinking
• Ask “help me think through” not “write me a PRD”
• Add specific examples and user quotes from research

Sub-agent feedback is surface-level

Cause: Agent personas aren’t specific enough, or PRD lacks depth

Fix:

• Enhance sub-agent personas with specific concerns (API rate limits, scalability)
• Provide agents with more context (@-mention technical/strategy docs)
• Ask for specific feedback types: “Focus on technical risks”
• Ensure your PRD draft has enough detail to review

Process takes too long

Cause: Too many steps, or not using parallelization

Fix:

• You don’t need all 3 approaches every time - one may be enough
• Use agents to generate drafts in parallel (not sequentially)
• Skip Socratic questions if you already have clarity
• Not every PRD needs sub-agent review - use for high-stakes features

Hard to track versions

Cause: Poor file naming or version management

Fix:

• Clear naming: feature-prd-v1.md, feature-prd-v2-chat-first.md, feature-prd-final.md
• Add dates: feature-prd-2025-01-15.md
• Keep changelog section in PRD
• Use git commits to track evolution
• Archive old versions in drafts/ folder

What’s Next?

You now understand how to partner with AI to write better PRDs: provide full context via @-mentions, use Socratic questioning to clarify thinking, generate multiple strategic approaches with agents, and get multi-perspective feedback via sub-agents.

Module 2.2: Learn about Data-Driven Decisions - use data to drive product decisions, from discovering funnel problems to estimating feature impact to analyzing A/B tests.

Interactive track: Type /start-2-2

About This Course

Created by Carl Vellotti. If you have any feedback about this module or the course overall, message me! I’m building a newsletter and community for PM builders, check out The Full Stack PM.

Source Repository: github.com/carlvellotti/claude-code-pm-course