By James Tang
Introduction
The role of a Product Manager (PM) has evolved. You are now expected to be a builder — someone who can quickly prototype ideas using AI, validate them with users, and then hand them off to engineers for production.
One of the most valuable skills you can learn today is using Large Language Models (LLMs) together with Cursor, an AI-powered code editor, to generate and iterate on code rapidly.
I recently completed the full AI Foundations and Coding Agents tutorials from Cursor (created by Lee Robinson). In this article, I’ll share a concise summary of the key concepts and a practical guide to using Cursor effectively as a PM.
Privacy First: Protecting Your Code
Before diving in, a quick but important note on privacy:
When working with AI coding tools, protecting your code and company IP is important. Most modern tools, including Cursor, give you the option to disable training on your data. I recommend always turning off machine learning training in Cursor (and any other AI tool) when handling sensitive or proprietary work.
Be cautious with tools like Gemini’s Design Stitch or similar features — many don’t offer an opt-out for training, which creates a risk of your work leaking to competitors.
Course Overview
The Cursor course is divided into two main parts:
1. AI Foundations
Covers core concepts:
- How AI models work
- Hallucinations and limitations
- Tokens and pricing
- Context management
- Tool calling and agents
2. Coding Agents
Focuses on practical application:
- Working with agents
- Understanding codebases
- Developing features
- Finding and fixing bugs
- Reviewing and testing code
- Customizing agents
- Putting everything together
Key AI Concepts You Should Understand
Probabilistic, Not Deterministic
AI models don’t always give the same output even with identical prompts. This is why it’s smart to test multiple models (GPT, Claude, etc.) in Cursor using the Cmd + K (or Ctrl + K) menu.
Hallucinations
Models can confidently invent things that don’t exist (e.g., suggesting a non-existent Tailwind vX package). This often happens because their knowledge has a cutoff date.
Tokens & Pricing
Everything is broken into tokens.
You pay per token, not per word.
Output tokens usually cost more than input tokens.
Context Management
Cursor maintains conversation history as context. The more you chat, the more context it has — but this also consumes tokens. Tool calling helps the model dynamically retrieve information instead of stuffing everything into context.
Tools & Agents
Cursor equips agents with tools such as: reading/writing files, searching code, running terminal commands, web search, and more.
Agents shine when given a clear objective. They can decide which tools to use, call them, analyse results, and iterate.
MCP (Model Context Protocol)
A new standard that allows AI models to use tools in a consistent, universal way.
End-to-End Example: Adding a New Dashboard Report
Here’s how this looks in real life. Imagine you just joined a SaaS company and need to add a new report to the dashboard.
Step 1: Understand the Codebase
Cursor automatically indexes your project. You can:
Use semantic search (understands intent and meaning)
Use instant grep (exact text/pattern matching)
Ask Cursor to generate Mermaid diagrams to visualise architecture, data flows, or component relationships.
Step 2: Create a Plan
Switch to Plan mode, describe the feature, and let the agent generate a detailed plan. Review and refine it before proceeding.
Step 3: Build
Click Build to let the agent implement the first version.
Step 4: Debug
For simple errors: paste the error message into Cursor.
For complex issues: use Debug mode. The agent forms hypotheses, adds instrumentation, asks you to reproduce the issue, analyses logs, and suggests targeted fixes.
Step 5: Review & Test
Ask the agent to:
Review the code
Write unit tests
Create Playwright integration tests
Break changes into commits for a clean PR
Pro Tip – Rules
You can create custom rules for the agent (e.g., coding conventions, preferred patterns, build/test commands). Use them sparingly — too many rules increase context and can confuse the model.
Cursor Marketplace
Don’t forget to explore the Cursor Marketplace — a hub for plugins and extensions that further enhance the AI’s capabilities.
Final Thoughts
As a PM, mastering tools like Cursor allows you to move from idea to validated prototype at lightning speed. You don’t need to become a full-stack engineer, but you can become dangerously effective at turning ideas into working software.
Start small, protect your IP, experiment with different models, and iterate quickly.
Reference:
Top comments (0)