Building with AI: The "Morning After"

By Philipp Marxen | Updated December 2025

Six months ago, in August 2025, we were drunk on the promise of AI.

We believed AI would democratize the web. We thought domain experts (beekeepers, bakers, lawyers) would replace junior developers. We coined the term "Vibe Coding"—the idea that you could just "feel" your way to an app by prompting, without understanding a line of code.

Now, in late 2025, the hangover has set in.

Tools are undeniably stronger than they were in early 2025. Cursor and Windsurf are now standard issue. Lovable and Bolt have moved from toys to production engines. But with great power comes a massive pile of technical debt.

The Good: Speed is no longer the bottleneck

The promise was partially fulfilled. You can now build a "Beekeeper Blog" or a "Shopify Extension" in 20 minutes.

• The IDEs: Cursor and Windsurf have effectively merged the role of "Architect" and "Typist."

• The Generators: Bolt and Lovable allow you to prompt entire full-stack applications.

The barrier to entry is zero. But the barrier to longevity has skyrocketed.

The Bad: The "Vibe Coder" hits the wall

In mid-2025, we celebrated the "Vibe Coder." This was the person who didn't know JavaScript but knew how to prompt.

By December, many of these Vibe Coders ran into the same wall: The Maintenance Cliff.

It turns out that building a feature is easy. Fixing a bug six months later in code you didn't write (and don't understand) is a nightmare. We are seeing a wave of "Zombie Apps"—projects that worked perfectly for the first demo but broke the moment an API changed or a library updated. Because the creator never understood the code, they can't prompt their way out of a complex dependency conflict.

The Ugly: Class Soup and Tech Debt

The specific ugly truth of 2025 is "Tailwind Class Soup."

AI models love Tailwind. They are trained on it. But they are terrible at organizing it. An AI will happily generate this for you:

HTML

<div class="flex flex-col md:flex-row p-4 m-2 bg-blue-500 hover:bg-blue-600 border border-gray-200 shadow-lg rounded-xl dark:bg-slate-800...">

It looks great. But when you ask the AI to "make the padding smaller on mobile," it might generate a slightly differentvariation for the next card. Suddenly, you have 50 cards with 50 slightly different padding classes. You have no system. You have no consistency. You have a "Frankenstein" frontend that is impossible to rebrand without rebuilding from scratch.

Webstudio Inception: The Professional's Firewall

This is where Webstudio Inception enters the chat—not as a magic wand, but as a firewall.

Inception is a frontend-only IDE. It allows you to use AI to generate that initial "Class Soup," but then it forces you to acknowledge it visually.

The Pro Workflow for 2026:

1. Generate with AI: Use Inception (or Cursor) to get the vibe right. Let it be messy.

2. Refine with Slots: Don't just paste the code. Identify the recurring parts (Navbars, Cards, Footers).

3. Tokenize: Instead of leaving bg-blue-500 hardcoded, map it to a Webstudio Variable (primary-color).

Will I use Inception to spin out quick projects? Yes. Will I use it to create long-term platforms? Only if I take the time to clean up the mess.

AI is no longer a replacement for the developer. It is a bulldozer. It moves earth fast, but if you don't know how to read the blueprint, you're just digging a very expensive hole.