Why AI is currently destroying one of the web's greatest certainties

There is an idea that built the modern web. It goes: Code is too complicated for normal people. So we build something in front of it. A layer that hides the complicated and shows the simple. Drag. Drop. Done.

This idea was right. It was right for years. It no longer is.

What an abstraction layer is and why we needed it

An abstraction layer is an intermediate layer between complexity and user. WordPress is an abstraction layer over PHP and database logic. Elementor is an abstraction layer over WordPress. Webflow is an abstraction layer over HTML, CSS and JavaScript. Shopify is an abstraction layer over e-commerce infrastructure.

These layers have a reason for existence: They make something powerful accessible without the user needing to understand the underlying complexity.

This worked. Over 43% of the entire internet runs on WordPress today (W3Techs, 2026). Millions of people have built websites they could never have built without these abstraction layers. This is no trivial result. This is a genuine democratizing effect.

The problem with abstraction layers

Every abstraction has costs. It hides complexity, but it also adds its own.

A page builder like Elementor stores layouts not as HTML, but as proprietary JSON in a WordPress database. Anyone who wants to leave this format struggles. Anyone who wants to debug it struggles more. Anyone who wants to let an AI work with it struggles the most, because models were trained on standard web formats, not on Elementor's internal data structures.

Even Webflow, which generates structurally cleaner code, remains a layer between user intent and result. This layer has its price: in flexibility, in speed, in the ability to work with modern tools.

Abstraction layers are worthwhile as long as the added value they create is higher than the costs they cause. That's the simple calculation behind it.

What AI does to this calculation

Andrej Karpathy coined the term "Vibe Coding" in February 2025: a workflow where users describe what they want in natural language, and an AI writes the code. Collins Dictionary officially named the term Word of the Year on November 6, 2025, and described it as a symbol of how deeply AI has already penetrated everyday life and language (Collins Dictionary, 2025).

This is not a fad. This is a structural shift.

Already 25% of startups in the Y-Combinator batch Winter 2025 had codebases that were 95% AI-generated (daily.dev, 2026). 65% of all developers worldwide use AI coding tools at least weekly (Stack Overflow Developer Survey, 2025, cited in MIT Technology Review, 2026). Platforms like Vercel and Netlify reported massive user growth from a completely new category: people who had never written code before and did it anyway with AI (The New Stack, 2025).

What's happening right now? The main reason for the existence of abstraction layers is disappearing. Code was inaccessible. AI makes it accessible. Not for everyone, not immediately, not without a learning curve. But systematically, recognizably, and in a clear direction.

The new calculation

If an AI can generate clean, semantic HTML, CSS and JavaScript directly from a natural language description, the question arises anew: Why do I still need an intermediate layer?

The answer before was: because without the intermediate layer I have no access to the result.

The answer today for a growing portion of users is: maybe I don't need it anymore.

This is not an absolute judgment. There are still users for whom a builder is the right choice, because they don't want to open a terminal, don't want to learn prompting, and know an existing system. This group is real and large. But it's shrinking, from top to bottom through the user pyramid.

Agencies and savvy freelancers migrate first. They recognize that the overhead of a proprietary builder system no longer outweighs the flexibility of AI-direct code. Technically somewhat savvy users follow. The rest remain longer, because network effects, habit, and ecosystem inertia are real.

What this means for the major platforms

WordPress builders like Elementor and Divi respond with AI features in the builder. This sounds pragmatic. It's structurally problematic: They force AI to work through proprietary intermediate formats instead of letting it loose directly on the target system. The AI becomes worse, not because it has become worse, but because the task is worse.

Webflow faces a different, more subtle problem. The platform has built the structurally cleaner model and generates real HTML/CSS instead of proprietary JSON. But its core promise, professional websites without programming knowledge, is attacked by the same AI wave that hits Elementor. Only from a different side and on a different timescale.

Both articles in this series examine these platforms in detail.

The actual thesis

Abstraction layers don't collapse because they were poorly built. They collapse because the premise on which they were built is changing.

Code was inaccessible. That was real. Builders were the answer. That was right.

AI is fundamentally changing access to code. This doesn't mean builders will disappear tomorrow. Network effects are enormous: WordPress alone powers over 63 million websites (W3Techs, 2026), Webflow counts 3.5 million registered users (ColorWhistle, 2026). These platforms aren't going away overnight.

But their strategic position is systematically weakening. The gap between what they deliver and what direct AI code can deliver is opening. Anyone making career or investment decisions around these platforms today should factor this in.

The abstraction collapse is not a thesis about tomorrow. It's an observation about a process that has already begun.

This text is part of a three-part series. Part 2: Why Elementor + AI is stillborn Part 3: Why Webflow solves the right problem wrong

Sources: