I Built Two AI Tools. The Second One Told Me How I Should Be Learning AI.

I tried the traditional way to learn AI. I opened an official course. I quit after a few lessons. The material was good but there were no stakes. Nothing broke. Nothing stuck.

I realized I learn best when things break.

Building TeachSim taught me LangGraph because the bot had to work. Building GitHub Digest taught me about silent failures because the code broke quietly. I learned the concepts because I needed them to function.

I want to learn AI engineering. I do not want to just use an API. I want to understand the tools I use every day.

Currently, my knowledge is surface level. I use slash commands and conventions to get things running. I do not understand MCP, subagents, hooks, or checkpoints. I just use whatever tool works that day.

I am changing my approach. I am building my own curriculum based on real projects. A phase is not done when I read about it. A phase is done when it fixes a real problem in my project.

Here is my new learning sequence:

• Phase 1: MCP (Extend what the agent can reach) • Phase 2: Skills (Formalize tool usage) • Phase 3: Subagents (Split work across multiple agents) • Phase 4: Hooks (Add guardrails) • Phase 5: Plugins (Package everything) • Phase 6: Checkpoints (Undo bad sessions) • Phase 7: Final Results

I will apply these to my live projects. For example, I will wire GitHubs MCP server into GitHub Digest instead of using a REST API.

This method takes more time now. I am trading immediate momentum for long term speed. I want to master the tools before I tackle heavy machine learning math. If I do not understand my tools, I will fight them while trying to learn hard concepts.

I am building this one step at a time.

Source: https://dev.to/mediblacksand_f0ea36c53fb/i-built-two-ai-tools-the-second-one-told-me-how-i-should-be-learning-ai-5el0

Optional learning community: https://t.me/GyaanSetuAi