The Ultimate Guide to AI Agents for PMs
Agents are the most valuable skill in AI and product. What is an AI agent? How to build an AI agent? Multi-agent systems, AI agent frameworks, tolls, and resources.
Hey, welcome to the premium edition of The Product Compass newsletter.
Every week, I share actionable tips, templates, resources, and insights for AI PMs.
Consider subscribing or upgrading your account for the full experience:
AI agents are the most valuable skill in AI and product right now.
This is an extended edition of several posts I recently published on social media with more new tools and resources.
In this issue, we cover:
What is an AI Agent
How to Build an AI Agent
🔒 What Are Multi-Agent Systems (With Agent Types and Architectures)
🔒Workflows, Agentic Workflows, and Agentic AI
🔒 AI Agent Frameworks: Comparison and Recommendation
🔒 Recommended AI Agent Tools and Resources
1. What is an AI Agent
Anthropic offered the best definition:
"Agents (...) are systems where LLMs dynamically direct their own processes and tool usage, maintaining control over how they accomplish tasks."
An AI agent is like an LLM on steroids. It can:
Autonomy: Pursue its goals autonomously.
Reasoning: Plan tasks, take action, analyze the results, and adjust.
Tools: Use tools (e.g., a function, MCP server, API, data retrieval, or code editor).
Memory: Remember its previous actions, or even learn from past interactions.
Some AI agents can also collaborate or delegate work to other AI agents or humans.
An example implementation of an AI agent with n8n with short-term memory, built-in tools (Google Sheets, Gmail) and Atlassian MCP server:
2. How to Build an AI Agent
Building the first AI agent might take just 30-60 minutes.
Here’s how:
Step 1: Define a System Prompt
It defines the goals, logic, and expectations.
I suggest you start with 11 AI Agent Prompting Principles: https://www.productcompass.pm/i/163255770/eleven-ai-agent-prompting-principles
Other guides:
GPT-4.1 Prompting Guide: https://cookbook.openai.com/examples/gpt4-1_prompting_guide
Anthropic Prompt Engineering: https://docs.anthropic.com/en/docs/build-with-claude/prompt-engineering/overview
Prompt Engineering by Google: https://cloud.google.com/discover/what-is-prompt-engineering?hl=en
Step 2: Select an LLM
Unless the framework handles iterating (e.g., n8n), start with a reasoning model (e.g., o1-mini).
Step 3: Connect Tools
What might your AI agent need to achieve its goals? Consider simple tools, like a calculator, custom functions, integrations, data sources, and MCP servers.
Step 4: Connect Memory
The agent must track its progress and learn. Most platforms support:
Short-term memory (variables, last interactions)
Long-term memory (vector, SQL, graph)
Step 5: Orchestrate the Logic
Whether a single agent or multiple agents working together, you must:
Map/code repeatable logic (flow) that doesn't belong to specific agents
Orchestrate communication between AI agents (static or dynamic)
You might also like the AI Agent Architectures With n8n Examples.
Step 6: Add User Interface
If your AI agent is user-facing, you can easily add logic using tools like Lovable, Bolt, or Google Firebase. No coding.
For more information, see:
Best Practices: How to Create SaaS Without Coding
A Complete Course: How to Build a Full-Stack App with Lovable (No-Coding)
Step 7: Evaluate the AI Agent
Rather than relying on standard metrics (hallucinations, helpfulness), perform error analysis and let metrics naturally emerge.
Our guides:
If your system involves RAG, evaluate retrieval and generation separately. Jason Liu’s There Are Only 6 RAG Evals provides a framework that maps well to this separation.
Bottom Line: Start Building. Stop Theorizing.
You can start with those step-by-step guides (<60 min):
Coming Soon: How to Build Anthropic Multi-Agent Research System
3. What Are Multi-Agent Systems (With Agent Types and Architectures)
Keep reading with a 7-day free trial
Subscribe to The Product Compass to keep reading this post and get 7 days of free access to the full post archives.