Learning a New Stack with AI: A Different Kind of Education

I've been a developer for over 20 years. In that time I've learned new technologies the traditional ways: technical books dog-eared and highlighted, video tutorial series watched at 1.5x speed, online courses with structured exercises and progress bars, and the slow accumulation of Stack Overflow answers bookmarked for later. Each of those approaches works. None of them work particularly well for an experienced developer who already knows a fair amount of what's being taught.

That changed with this project.

What I Already Knew

Before starting this rebuild of heathschweitzer.com, I had used AI tools in development in smaller ways. I leveraged GitHub Copilot chat to think through a few features on a WordPress build in 2025. I completed an AI bootcamp through a former employer that covered LLM fundamentals and put Python and Azure infrastructure to work in practice exercises. I understood conceptually what these tools could do.

But this was the first time I used Claude Code, the first time I integrated the Anthropic API directly into a project, and the first time I built a development workflow specifically designed around AI collaboration from day one. The difference in scale and depth compared to those earlier experiences was significant.

Skipping What You Already Know

The most valuable thing about AI-assisted learning for an experienced developer isn't what it teaches you. It's what it lets you skip.

When I started learning Next.js, I didn't need anyone to explain what a database is, what a foreign key constraint means, or why you should never store credentials in source code. I've known those things for two decades. Traditional learning resources — courses, books, tutorials — have to explain them anyway because they're written for a general audience.

With Claude as a collaborator, I could say "I've been building on LAMP for 20 years, I understand relational databases and server-side rendering, explain Next.js Server Components in terms of what's different from PHP, not from scratch." And I got exactly that. The explanation started from my existing mental model and bridged to the new concept. No preamble, no "first, let's understand what a web server is."

That compression — jumping straight to the gap between what I know and what I need to know — is something no pre-recorded course can do. The curriculum is fixed. The AI adapts in real time.

Learning While Doing Real Work

The other shift is the nature of the work itself. Most learning resources have you build something artificial — a todo app, a weather widget, a blog with fake data. The exercises exist to teach the concept, not to produce anything useful.

This project was different from day one. I was rebuilding my actual personal site, migrating real content from a WordPress installation that had been running since 2012, deploying to a real server I already owned, and publishing posts people could actually read. Every feature we built solved a real problem. Every bug we fixed had a real consequence.

That stakes-based learning sticks differently. When the WordPress importer failed to parse the XML correctly, I wasn't debugging a toy exercise — I was trying to recover ten years of writing. When the production deployment threw a 502, my site was actually down. The urgency made every lesson land harder.

The Workflow

What emerged over the course of this project was a genuine development workflow, not just occasional AI assistance:

Claude.ai (this interface) for architecture decisions, learning discussions, and understanding why things work the way they do. When I hit a concept I didn't fully grasp — React Server Components, the difference between publishedAt and updatedAt, how NextAuth handles JWT sessions behind a reverse proxy — I asked here and got explanations calibrated to my background.

Claude Code for implementation. Once a feature was understood and spec'd, Claude Code made the actual file changes. I reviewed the diffs, tested locally, and committed what worked. The iteration cycle — spec → implement → test → refine — happened in minutes rather than hours.

My own judgment for everything else. Which technology choices made sense for my constraints. Which features were worth building now versus later. When the AI's suggestion was technically correct but wrong for my situation. That judgment is the part two decades of experience actually trains.

The AI didn't replace my development skills. It amplified them by handling the parts that require knowing syntax and API surface area while I focused on the parts that require experience and judgment.

What This Means for How We Learn

I keep thinking about the traditional e-learning industry — SCORM compliant courses, the bootcamps, the certification programs. Their model is built on a fixed curriculum delivered uniformly to every learner. You progress through the material at roughly the same pace as everyone else, regardless of what you already know.

That model has a ceiling. It can only compress learning so much before you hit the constraint of "this lesson requires the previous lesson." The curriculum is linear because teaching is linear.

AI-assisted learning isn't linear. It meets you where you are, adjusts to what you already know, and lets you move at the speed of your understanding rather than the speed of the course. An experienced developer can skip five modules of foundation material and go straight to the advanced concepts. A beginner can ask the same question five different ways until it clicks, without feeling self-conscious about it.

I don't know how traditional e-learning companies will compete with this if they don't fundamentally rethink their model. The value was never the content — content is now abundant and free. The value was the structure and the guidance. AI provides better structure and better guidance, personalized to each learner in real time.

The ones who survive will be the ones who figure out how to wrap AI into a learning experience that's more than just "here's Claude, good luck." Curation, accountability, community, credentialing — those still matter. But the lecture-watch-quiz-repeat model is going to struggle.

What's Next

This is the second post in an ongoing series about rebuilding heathschweitzer.com on a modern stack. The first post covered the technical journey — what I built, the mental model shifts, and how the deployment works.

This one is about how I learned it.

The next posts will cover specific features as we build them — AI-assisted content generation in the post editor, category management, and eventually the design polish pass that will make the site look as good as it functions.

If you're an experienced developer sitting on a skill you've been meaning to learn, I'd genuinely encourage trying this approach. Pick a real project with real stakes. Use AI as a collaborator, not just a search engine. Skip the parts you already know. The learning curve is real — but it's shorter than you think, and the work you produce along the way is actually useful.