Tutorial-driven development with Kiro

February 8, 2026

During the week, I teach developers about Kiro’s spec-driven development. On the weekends, though, I’ve been doing something different: tutorial-driven development. I ask Kiro to create a targeted tutorial just for me.

Yes, there are existing tutorials on just about every topic. But AI lets me build one tailored to my interests and skill level. I get to build something I actually care about while I learn. That’s just more engaging than following a generic walkthrough.

I can also customize the tutorial to my needs. If I already know how to do something in one language but want to learn it in another, the tutorial can make those comparisons for me. If I want to skip the basics and dive into advanced topics, it adjusts. The tutorial meets me where I am.

Learning FastMCP with a Family Tree Server

Genealogy has always been an interest of mine. Yeah, I know—I’m too young for that hobby. But here we are.

I wanted to learn FastMCP, the Python library for building Model Context Protocol servers. Rather than following a generic tutorial, I asked Kiro to create one around something I’d actually use:

1
2
3
4
5
6
7
8


I want to create a Family Tree MCP server using FastMCP. The MCP server 
should have tools for finding and listing individuals, families, facts, 
sources, etc. Let's use a simple GEDCOM file as the source. I want you 
to create a tutorial that guides me through building the application. 
Your goal is to teach me about the various aspects of FastMCP. The 
application is secondary to learning. The tutorial should start small 
and then add concepts to build my understanding. Break the tutorial 
into modules.

Kiro generated a three-part tutorial that walked me through GEDCOM parsing, the four MCP primitives (Tools, Resources, Resource Templates, and Prompts), and finally deploying to AWS with AgentCore Runtime.

The result? A working MCP server loaded right into Kiro. I don’t actually do genealogy research in my IDE, but it’s a great place to test the tools before connecting them to an agent.

Building a Genealogist Agent with Strands and AgentCore

With the MCP server working, I wanted to go further. I wanted to learn Strands Agents and AgentCore—how to build an AI agent that could use those tools, deploy it to the cloud, and add production features like memory and access control.

Same approach:

1
2
3
4
5
6
7
8
9


I want to create a genealogist agent with Strands and AgentCore. This 
agent should guide me in my research and make suggestions along the way. 
As with the Family Tree MCP server, I do not want you to build this. I 
want you to create a tutorial that guides me through building the 
application. Your goal is to teach me about the various aspects of 
Strands and AgentCore. The agent is secondary to learning. The tutorial 
should start small and then add concepts to build my understanding. 
Start with a local Strands agent, then AgentCore Runtime, Identity, 
Gateway, Memory, Policy, etc. broken into modules.

Kiro created a seven-module tutorial that progressed from a three-line “Hello Strands” agent all the way to Cedar policy enforcement through an AgentCore Gateway.

The agent connects to my family tree MCP server and actually helps with research. It spots anomalies in the data (“Mary would have been 11 at her first child’s birth—that date might be wrong”), suggests records to look for based on historical context, and remembers our conversation across sessions.

Why This Works

The magic isn’t that AI wrote a tutorial. It’s that AI wrote my tutorial.

Generic tutorials optimize for the broadest audience. They explain things I already know and skip things I need. They use contrived examples that don’t stick. Tutorial-driven development flips this: I describe what I want to learn and what I want to build, and the tutorial adapts to both.

A few things that made this effective:

Real data, real motivation. Using my actual family tree meant I cared about the output. When the agent found an inconsistency in my great-grandmother’s records, I wanted to fix it.
Progressive complexity. Each module built on the last. I wasn’t overwhelmed with AgentCore deployment details before I understood the basic agent loop.
Immediate feedback. I could test each step in Kiro before moving on. When something didn’t work, I asked Kiro to help debug it—still in the context of the tutorial.
Tailored explanations. When I asked for comparisons to patterns I already knew, I got them. When I wanted to skip ahead, I could.

Try It Yourself

Next time you want to learn a new framework or library, try this: don’t search for an existing tutorial. Ask your AI assistant to create one for you. Tell it what you want to build, what you already know, and how you learn best.

You might be surprised how much more you retain when the tutorial is built around something you actually care about.