The shrinking gap – why I left Webflow and rebuilt with React and AI

I built the Webflow site. I should say that upfront because this isn’t a takedown from the outside. I knew the tool well enough to build a polished actress portfolio in it – responsive, fast, good typography, working CMS. It did the job.

Then Webflow stopped being Webflow.

In April 2024 they acquired Intellimize, a conversion rate optimisation startup. The CEO stepped down. The new leadership pivoted to something called a “Website Experience Platform” – Bayesian neural networks, firmographic deanonymisation, enterprise CMOs. The GUI hasn’t meaningfully changed since 2015.

They looked at AI commoditising the creation layer and decided to abandon it. Went upmarket. Left designers holding a tool that’s getting more expensive and less loved with every quarterly roadmap. A smart business move, probably. But it left me in a room where the lights were being turned off.

The gap you stop noticing

Visual builders don’t shrink the distance between your idea and the finished product. They replace it with a different distance – between what the platform can express and what the work actually needs.

For years the trade was worth it. The platform gap was smaller than the code gap. You could ship 80% of your vision in a fifth of the time. The 20% you left behind – the easing that wasn’t quite right, the content that couldn’t adapt, the interaction that almost worked – was the cost of speed.

That cost used to be invisible. Once you notice it, you can’t stop noticing it.

What the extra 20% actually looks like

I started rebuilding on a quiet weekend. No plan, no brief. Just a laptop and Claude open in the terminal.

The first thing I built was the showreel. In Webflow, it was a poster image with a play button. Standard. Nobody complains about this because every portfolio does it.

But think about what happens in the ten seconds after a casting director lands on the page. They’re triaging. 200 tabs open. They’re not going to click play on a speculative bet. They need to see quality before they decide to invest attention. A still image asks them to take your word for it. A moving image shows them.

So the reel autoplays. Silent, looping, immediate. The casting director is already watching before they’ve decided to watch. The assessment of lighting, framing, performance quality happens before the rational brain catches up. By the time they reach for the volume, the first gate is already passed.

This is a five-minute change in code. It’s an impossible change in Webflow – not because Webflow can’t autoplay video, but because the specific control you need requires expressing precise intent, not selecting from a menu of approximations.

Context-aware content

The size card sits above the credits, not below them. A casting director scanning for a specific physical type doesn’t care about credits yet. They want height, build, look – in the units they think in.

The site reads your timezone. America gets 5’7” and 120 lbs. Europe gets 170 cm and 54 kg. A segmented control lets you switch manually, but the default is already right for wherever you are.

A casting director in Prague doesn’t pause to convert centimetres. The flow doesn’t break.

That’s the kind of decision that doesn’t occur to you when you’re working inside platform constraints. And even if it did, the implementation is a custom script injected before </body>, disconnected from the component it serves, easy to lose track of, impossible to reason about six months later. The tool’s vocabulary doesn’t include it.

The filmmaking page

A production still fills the viewport. As you scroll, it fades. Blur creeps in. Colour drains. Brightness drops. Credits emerge over the darkening image like end titles over a final frame. At the bottom, everything is black, and you’re in the gallery.

The page behaves like a film because it’s a page about films. The medium matches the message. In a visual builder, this is either a plugin or a compromise.

The content update that proved the approach

Olga sent a document. 500+ lines. Every credit restructured. New categories. New representation – management in LA, agents in London and Warsaw.

I described the structure to Claude. We updated everything in parallel – content files, components, page layouts. New sections appeared. Old ones reorganised. The locale switcher emerged from the data itself – measurements that varied by region suggested an interface that responds to region. Training entries revealed a narrative arc that suggested a visual hierarchy.

Content didn’t fill a template. Content generated design decisions. That’s what happens when the container can reshape itself as fast as the material arrives.

In Webflow, this is a rebuild. New CMS collections. Reconnected dynamic lists. Copy-pasting 500 lines of text into modals three levels deep in the GUI, one field at a time, for three hours. Instead, the whole update took ten minutes – because the content lives in a file, not behind a form.

The footer that does persuasion work

The old footer said “Let’s talk” with an email button. The new one lists management, UK agent, Poland agent – named contacts, real companies, tiered by how the industry actually works. A casting director reads it in two seconds and knows exactly who to call.

What actually changed

Webflow abandoned the creation layer to sell conversion optimisation to enterprises. Fair enough. They don’t need me. But I don’t need them either.

I can tune page speed, harden security headers, optimise for AI search engines, and ship a content update – all in the same afternoon. No platform middleman. No waiting for a feature request to make it onto someone else’s roadmap. The creation layer didn’t lose value. It lost friction.

The gap between idea and execution used to be wide enough that tools like Webflow were the obvious bridge. That gap is shrinking. And as it shrinks, the interesting work moves from “how do I build this?” to “what should I build?”

That’s not a tool problem anymore. That’s a design problem. And design problems are the ones worth having.