Why is it so difficult to go from learning AI to actually building AI applications?

I've been working on a learning platform to address a problem I've experienced while exploring AI development: there are plenty of tutorials, documentation pages, and individual courses, but figuring out what to learn first, how the pieces connect, and what to build next can be overwhelming.

I'm organizing a curriculum that starts with programming fundamentals and progresses through APIs, databases, machine learning, LLMs, RAG, tool calling, LangChain, LangGraph, MCP, fine-tuning, n8n, and multi-agent systems. The platform is being designed around 3,200+ practice exercises, hundreds of practice projects, interview questions, and portfolio projects, with a path from beginner concepts to advanced topics.

One thing I'm trying to get right is the progression: learners shouldn't just recognize terminology; they should be able to apply concepts and build things independently.

For people learning or working in AI, what helped you make the jump from following tutorials to building your own applications?
Was it structured exercises, larger projects, debugging real problems, documentation, or something else? I'd especially appreciate hearing what you found missing from existing learning resources.