My First 365 Days
of Vibe Coding

The first thing I did was hire my friend Ognen — a senior developer — to mentor me. Not to build for me, but to teach me how to think like a developer instead of a marketer.

He helped me refine the tech stack I'd planned out in my long conversations with o3, taught me how to organize projects so the AI could actually help, and showed me the debugging commands that changed everything: tsc and npm run lint. Before that, I was running the build over and over, fixing one error at a time. These two commands let me see all the problems at once. That moment became the foundation for our Debugging Flow lesson — still one of the most valuable lessons in the Builder Codex.

Then came the information binge. YouTube became my university. Databases, authentication, authorization, payment processing — concepts I'd heard of as a user but never had to manage myself.

My first tool was v0 — a simple AI builder that gave me a taste of what was possible. But I quickly moved to Cursor. The playground tools are great for experimentation, but if you're building something real — something you'll maintain and scale — you need to understand your project architecture and be able to integrate any external resource that enhances your product.

March 29th, 2025 — my 43rd birthday. Goal: 1 paying customer. Result: 7.

The product was CaptAIn. The idea was simple: I was annoyed by jumping between AI chats every week — ChatGPT, then Gemini, then Claude — and having to repeat my business context every time. So I built a tool to save my context and switch between whichever model was currently best.

That moment — seeing real people pay real money for something I built — changed my identity. "I can't build" became "I just built something, and real people actually wanted it."

But I also realized I needed to talk to people. I had Ognen for the developer perspective, but I wanted to connect with others who were coming at this from the non-developer side.

So Christina and I launched Vibe Coding Cologne — an in-person meetup for people learning to build with AI. We hosted it at Startplatz, our co-working home for eight years, with incredible support from the Graef family — Lorenz and his three sons. We created a WhatsApp group that quickly grew to over 500 members — a place where I'd never miss another significant AI model launch or new tool. If you want to join the conversation, the Vibe Coders group is open.

I'm incredibly grateful for everyone who's contributed to that community. It kept me sane during the hard parts.

Among many other things, one of the most valuable things I discovered through the community was the power of MCP servers. Think of them as arms and legs you give to the brain — the large language model — so it can reach beyond the codebase and interact directly with the services you're using. These became essential to my workflow:

By Q3, I'd built enough to finally see the journey — where I'd been and where I was heading. The path that felt like chaos in the moment now looked like a winding road up a mountain, each struggle a switchback that got me higher.

With that clarity came the urge to codify — to turn hard-won lessons into reusable systems. This has been a habit since my youth. I was a student tutor throughout school and university, and teaching always helped me solidify my own understanding. Once I could explain a concept in simpler terms, it stuck. Pirate Skills, at its essence, is that same instinct: a platform that helps me deepen my understanding of building and growing products by teaching others.

First came the Builder Codex — a clear 6-step structure for setting up any new app. I had community members and experienced developers like Ognen and Gero, and many other technical and non-technical founders helped me make improvements. We set it up for free on pirateskills.com, built entirely with vibe coding techniques.

Then I wanted to build my next project with others — to teach what I'd learned while I was still learning. The idea was Signal Log: a voice-first project journal for indie founders. The concept was simple — drop quick spoken check-ins that become tidy progress logs. An AI coach reflects back what you've done, challenges your next step, and keeps a gentle cadence across all your ideas.

I recruited a pilot cohort from the Vibe Coding Cologne community to build it together. Every week, we'd work on Signal Log as a group — me teaching what I'd just learned, them building alongside me.

That experimental group became the first cohort of what's now the Builder Forge — a 6-week program that runs every quarter for people who want to build real products.

After months of heavy building focus, I realized the pendulum had swung too far. I wasn't applying my 15 years of growth experience to what I was building.

So I built the Growth Codex — everything I'd learned about marketing, sales, and growth, codified into a system. Also free on pirateskills.com.

Writing the Growth Codex made me realize how much on track I'd already been in Q2 with our mantra of Build. Grow. Repeat. But of course, this has its price. Here's how my AI spending developed — for both building and growing my fledgling SaaS empire:

The chart tells a story. I started at the beginning of 2025 with just the $20 Cursor Pro subscription — learning, fumbling, not using many tokens at all. By April, my usage climbed as my skills improved.

Then I got scared. I saw the costs rising and thought: "Am I just going to keep spending more and more?" So I pulled back for two months.

In July, I had a realization: fuck it — doesn't matter. The value I was getting from AI was so much higher than the cost. I upgraded my subscription and leaned in. The way I see it now: a $200 subscription is like having three pre-AI developers at your fingernails. Instead of paying $500 a day, you're paying $10. Of course, this means I need to be much more involved in the product development process. But that actually turned out to be an incredible speed boost — no more daily or weekly meetings with other people, which adds even more to the value. This could be absolute copium, and I'm curious to see what I'll say in a year.

October was my peak month — over $300. I'd embraced Claude Opus 4.5 for complex tasks, and it felt like gaining superpowers. But I also got smarter about token usage. By November, I'd learned when to use the expensive models and when Cursor's Composer 1 would do.

Claude Opus 4.5 felt like a breakthrough — not just for me, but for the entire vibe coding community. Even the previously skeptical senior developers started to realize that with this model, the game had truly changed.

Recently, I've been using Claude Code within Cursor. Anthropic's philosophy resonates with me, and they seem to be at the cutting edge of what matters — MCP servers, skills, the infrastructure that makes AI-assisted building actually work. January 2026 is an experiment: can I get most of what I need from the $90 Claude Code Max subscription plus $20 for Cursor? We'll see next month.

Day 365: I can build and grow things I couldn't have imagined a year ago. I'm still learning fundamental software and product development principles every week. But I ship — and that's what matters.

If I could go back to Christmas 2024 and give myself five pieces of advice, here's what I'd say: