TL;DR

When you stop building for yourself and start building for others, everything changes

The projects I described in Part 2 — the VAT validator, the trading system, the Wochenmarkt app — were all useful. Things I built because I needed them or because someone near me needed them. CalendarGenie and Post-er changed that.

When you stop building for yourself and start building for others, everything changes

The projects I described in Part 2 — the VAT validator, the trading system, the Wochenmarkt app — were all useful. But they were tools. Things I built because I needed them or because someone near me needed them. They weren’t products.

CalendarGenie and Post-er changed that. These were the first things I built that strangers on the internet found, used, and came back to. And that shift — from builder to product maker — taught me more about software than any coding tutorial ever could.

CalendarGenie: The Simplest Idea That Actually Works

The idea came from a moment of pure frustration. Someone sent me a screenshot of a conference schedule — you know the type: a dense PDF page with dates, times, session titles, all formatted beautifully for print and terribly for anything digital.

I wanted those events in my calendar. Not five of them. All forty of them.

The manual way: read each line, open calendar app, create event, enter title, enter time, save. Repeat forty times. Lose will to live around event seventeen.

So I built CalendarGenie. You upload a screenshot — any screenshot — and AI extracts every event it can find. Dates, times, titles, locations. It gives you a clean list, you pick what you want, and it exports to your calendar. Done.

The technical implementation is straightforward: the image goes to a vision model (GPT-4o at the time, now configurable), which returns structured event data. The app parses it, presents it for editing, and generates .ics files or direct calendar integration.

What surprised me was how universal the use case turned out to be. Conference schedules, yes. But also:

  • University lecture timetables (students were early adopters)
  • Sports season fixtures
  • Work shift rosters shared as screenshots
  • Event flyers from social media
  • School holiday calendars

The lesson? The best product ideas are the ones where you build for yourself first, then discover the world has the same problem. I didn’t do market research. I didn’t build a business plan. I scratched my own itch, and it turned out millions of people have the same itch.

CalendarGenie still runs on Replit. It’s lean, it’s fast, and it does exactly one thing well. That last part matters more than I initially understood.

Post-er: When Ambition Scales Up

If CalendarGenie was „solve one problem perfectly,“ Post-er was „what if I tried to solve a whole workflow?“

The problem: creating professional social media content is tedious. You need a good image, a compelling caption, proper hashtags, maybe different versions for different platforms. If you’re managing a brand or a small business, this eats hours every week.

Post-er started simple — generate a social media post with AI. But it grew. And grew. And the fact that it grew is both the best and most dangerous thing about vibe-coding.

Here’s what Post-er does today:

Five AI Image Providers

This was born from frustration with quality inconsistency. DALL-E would nail one prompt and botch the next. Midjourney-style results were different from Stable Diffusion-style results. So instead of picking one, I integrated five:

  • DALL-E 3 (via OpenAI) — great for clean, commercial imagery
  • Stable Diffusion (via Stability AI) — flexible, artistic
  • Flux (via Replicate) — photorealistic, newer model
  • Ideogram — strong with text in images (logos, signs)
  • Recraft — design-focused, excellent for brand content

Users can generate from all five simultaneously and pick the best result. Or they can stick with their favorite. The point is choice — because different use cases need different aesthetics.

Veo3 Video Generation

This one still feels like magic to me. Google’s Veo3 video model, integrated right into the post creation flow. Describe a scene, and you get a short video clip you can use for social content.

Is it perfect? No. Video generation in 2025 is still hit-or-miss. But when it hits — a 5-second product showcase, a moody brand clip, an attention-grabbing story opener — it’s genuinely impressive. And doing it from within the same tool where you created your image and wrote your caption? That’s a workflow that didn’t exist before.

The Credit System

This is where I learned about building for sustainability. AI image generation isn’t free. Every API call costs money. And if you’re offering a product to users, you need a model that works.

I built a credit system: users buy credit packages, and each generation costs credits. Different models cost different amounts (Veo3 video costs more than a DALL-E image, naturally). Users can see their balance, track usage, and top up when needed.

It’s not glamorous. It’s plumbing. But it’s the plumbing that separates a demo from a product. Without a credit system, Post-er is a cool toy that bankrupts me. With one, it’s a sustainable tool.

PWA: Install It Like an App

Post-er is a Progressive Web App. You can install it on your phone or desktop, and it behaves like a native app — home screen icon, full-screen mode, offline awareness.

This was another lesson in thinking like a product person rather than a builder. Users don’t care about your tech stack. They care whether they can open the thing quickly and use it while commuting. Making Post-er a PWA tripled mobile usage almost overnight.

The Difference Between Building and Product-Making

CalendarGenie and Post-er taught me something the earlier projects hadn’t: building software and building a product are fundamentally different skills.

Building software is about making something work. Does the button do the thing? Does the data save? Does it not crash?

Building a product is about making something useful, usable, and sustainable. It’s about:

  • Onboarding: Can someone who’s never seen your app figure it out in 30 seconds?
  • Error handling: When something goes wrong (and it will), does the user know what happened?
  • Economics: Can you afford to run this thing? Does the pricing make sense?
  • Retention: Will they come back tomorrow? Next week?

None of these are coding problems. They’re product problems. And ironically, my 25 years of non-coding product experience turned out to be the most valuable asset I had.

A professional developer building Post-er would write cleaner code. But would they integrate five image providers because they felt the user frustration of inconsistent outputs? Would they build a credit system in the first version because they’d lived through the „cool demo that costs too much to run“ cycle at DATAlovers? Would they make it a PWA because they’d watched people in Bitkom meetings struggle with bookmarks?

This is what I mean when I say non-developers see problems differently. My lack of engineering purity is compensated by two decades of watching real people use real products.

What Both Projects Taught Me

CalendarGenie taught me the power of simplicity. One input, one output, zero confusion. The best apps I’ve used in my life all follow this pattern. CalendarGenie doesn’t try to be a calendar app, a scheduling tool, or a productivity suite. It turns screenshots into events. That’s it. And that focus is exactly why it works.

Post-er taught me the cost of ambition. Five image providers means five API integrations, five sets of error handling, five billing relationships. Video generation means dealing with long processing times, user expectations, and failed generations. A credit system means building payment flows, handling edge cases, and thinking about fraud.

Every feature you add is a commitment. Not just to build it, but to maintain it, support it, and pay for it. Vibe-coding makes building fast. It doesn’t make maintaining fast.

If I built Post-er today, knowing what I know now, I’d launch with two image providers and add the others later. I’d skip video generation for v1. I’d validate the credit system with a manual process before automating it.

But I didn’t know that then. And honestly? The messy, ambitious, over-featured journey is how I learned.

Still Running, Still Evolving

Both apps are live today. CalendarGenie at [calendargenie.replit.app](https://calendargenie.replit.app). Post-er at [post-er.replit.app](https://post-er.replit.app). Both still on Replit, both still serving real users.

They’re also both at the limits of what Replit can comfortably handle. Post-er especially — with its multiple provider integrations, background video processing, and credit management — is pushing against the platform’s constraints.

Which is part of why I made the jump to more powerful tools. But before I talk about the products that came next, I need to talk about the tools that made them possible.

This is Part 3 of a 7-part series about my journey from zero coding experience to building production applications with AI. In Part 4, I’ll talk about the tooling upgrade that changed everything: Claude Desktop and Claude Code.

Try what I’ve built:

  • 🧞 [CalendarGenie](https://calendargenie.replit.app) — Turn any screenshot into calendar events
  • ✍️ [Post-er](https://post-er.replit.app) — AI-powered content creation with 5 AI providers and video
  • ⚡ [Energy Backbone](https://energy-backbone.bigdataheaven-software.de) — Smart energy management
  • ⚓ [Marina Master](https://marina-master.bigdataheaven-software.de) — AI-powered marina management