A PM’s Guide to Using AI

… just a quick hit to provide working links over the holidays. Please excuse the roughness of the design.

1. Getting Started

The first thing is to understand AI PM is just a subset of PM. Many of the tasks are the same, but you’re building AI features or products, and using AI tools even more.

AI Product Management Basics:

Prompt Engineering
Context Engineering (RAG)
AI Evals & Testing
AI Prototyping/Vibe Coding
Agent Workflows

Key Differences from Traditional PM:

Technical depth required
Building AI features + products
Higher expectation of AI tool use

Deep Resources:

2. Prompt Engineering

It’s the #1 skill for great agents + productivity.

Free Guides:

Techniques Grid:

COT	ROLES	EXAMPLES
Few-Shot	Constraints	XML Tags
Chain Prompts	Autonomy	Artifacts
Step-by-Step	Persist	Self-Consistency

Tools:

Anthropic Claude
ChatGPT
ChatPRD

3. Context Engineering & RAG

Vector Databases:

Pinecone
Chromatic
Qdrant

Knowledge Graphs:

Neo4j
GraphRAG

Frameworks:

LlamaIndex
LangChain

RAG vs Fine-Tuning vs Prompting:

Use prompting for simple instructions and general tasks
Use RAG when you need to retrieve specific knowledge or documents
Use fine-tuning for specialized domains with consistent style/format needs

Context Engineering Guide: Step-by-Step

4. AI Prototyping & Vibe Coding

No-code first:

Lovable
Bolt
v0
Replit Agent
Devin

IDE first:

Cursor
Windsurf
GitHub Copilot
Codeium

Others:

Jules
Codex

Infrastructure Tools:

Supabase
Firebase
Clerk
Netlify
Vercel
Railway
DigitalOcean
GitHub

Guides:

5. AI Agents & Agentic Workflows

Agent Platforms:

n8n
Crew AI
Zapier
AutogenAI
Make
LlamaIndex
LangGraph
LangChain
Lindy
Lamini
Haystack
Cassidy
FlowWise

Techniques:

Tool use
MCP
Agent Architectures
ReAct
A2A RAG

Guides:

6. AI Evals, Testing & Observability

Evaluation Platforms:

X (Twitter)
Arize
Braintrust
Weights & Biases

Testing Approaches:

Unit Tests → LLM Judge/Error Analysis → TNF/TPR
Human Eval → Model Eval Train/Test/Dev → A/B Tests

Key Metrics:

Accuracy / Precision / Recall
Latency (P50, P95, P99)
Cost per request
User satisfaction

Virtuous Cycle Diagram: Build → Evaluate → Iterate → Observe → (repeat)

Resources:

7. Foundation Models

Leading Models:

Best reasoning & coding: Claude (Anthropic)
Most versatile: GPT-4
Multimodal leader: Gemini
Real-time data: Perplexity
Long context: Claude
Efficient: Gemini Flash
Open source: Llama
Multilingual: Command R+

Model Types:

LLM (Large Language Models)
LCM (Large Code Models)
LAM (Large Action Models)
MoE VLM (Mixture of Experts Vision-Language Models)
SLM (Small Language Models)
MLM (Multimodal Language Models)
SAM (Segment Anything Models)

8. AI PRDs & Building

Essential Frameworks:

AI Product Strategy Template
AI PRD Structure
Evaluation Framework
Risk Assessment Matrix

Free Templates:

Common Mistakes:

No fallback strategy
Ignoring latency/cost
Building without evals
Overfitting to demos

Key Resource:

Most People Build AI Products Wrong – How to Do It Right

9. Career Resources

Breaking Into AI PM:

Build your AI PM background
Create an AI PM Portfolio
Update your Resume + LinkedIn
Get Referrals to AI PM Jobs
Be Over-Prepared for Unique Interviews:
- a. Vibe Coding
- b. AI ethics + guardrails

Key Guides:

Tools & Communities: