I gave up on visual editors

For years I thought the visual editor was the holy grail for no-code A/B testing. Click on a headline, change the text, ship a variant without touching code. That was the dream.

And honestly, it worked… until it didn’t.

As sites became more complex, the visual editor started breaking in all the ways growth teams already complain about:

1. Simple changes weren’t actually simple

Modern websites are built with nested components, dynamic rendering, responsive layouts, and all sorts of weird edge cases. The visual editor often grabbed the wrong element or broke something on mobile. You’d fix one thing, and something else would glitch.

2. The “no-code” promise quietly required… learning CSS

The moment the editor couldn’t correctly identify an element, it would ask for a CSS selector. At that point, most non-technical users would freeze. It stopped being “simple,” even if the interface still looked friendly.

3. Meaningful experiments became fragile

Small cosmetic changes were fine. But anything involving pricing components, layouts, modals, or dynamic content became a headache. The editor just wasn’t built for real product-level experimentation.

After fighting this for long enough, I finally accepted that the visual editor wasn’t the future. It was a clever workaround for an older era of the web.

So I built something else: Generative Experimentation (GX)

Instead of clicking around on a page and hoping the editor guesses the right element, you just describe the change you want in plain language.

Example:

“Please help me reposition this side thumbnail selection to the bottom of the main product image.”

Under the hood, the system:

1. Understands the intent
2. Finds the correct elements in the live DOM
3. Generates the code
4. Previews it safely before launch

No dragging boxes. No CSS selectors. No fear of breaking the layout.

It basically shifts A/B testing from “manually editing the DOM” to “expressing the idea and letting the system figure out the implementation.”

What surprised me most

People move faster. Developers stop doing low-leverage testing tasks. Marketers aren’t blocked. And experiments that used to take a day now take a few minutes.

Honestly, I didn’t expect this to feel like such a big shift… but it kind of is.

Curious what IH thinks

Are you also seeing the limits of visual editors?
Would you use a plain-English way of generating experiments?
Anything you’d expect a tool like this to handle that others don’t?

If you’re curious what I built, here’s the product page (not trying to hard-sell, just sharing): https://www.mida.so/generative-experimentation