In one weekend, AI rebuilt our entire website.
515 pages. 4.8 million users. A $3M ARR business. We moved all of it off Webflow, with Claude Opus 5.5 doing the heavy lifting.
Nobody noticed. That was the point.
The site is now faster (our desktop PageSpeed score went from 38 to 99), cheaper to run and fully ours. And for the first time, any engineer on our team can ship a change to it, not just a Webflow specialist.
Here's why we left, how we did it without breaking anything, and what I'd tell any team thinking about doing the same.
Why we left Webflow #
Last year we finished a complete rebuild and redesign of Rezi.ai on Webflow. It took several months and a real investment to get the quality where we wanted it.
Not long after, AI models got dramatically better, and so did the workflows around them. We started asking how we could put them to work on our own website.
On Webflow, every change followed the same path: designer, then Webflow developer, then production. Our whole funnel lived inside Webflow, so every change went through that chain. The bottleneck was always there, but as AI sped up everything else, it got tighter. Eventually the opportunity cost of not moving was bigger than the cost of moving, and we were confident it was the right call.
Moving to code opens up a lot more:
- Any engineer on the team can contribute, not just a Webflow specialist
- What we can build, without the limits of a visual editor
- Pulling other data sources straight into the site
- A chance to rethink the blog and how we handle and work with content from an SEO perspective
- Technical fixes we couldn't make before
- Sweeping changes across the whole site, instantly
- Harnesses that make sure our rules are followed, quality is controlled and the site stays consistent
The final push was a price quote. We wanted to add one language to the site. Webflow wanted $2,500 a year for it. One language.
That landed on top of a three-year enterprise contract worth about $30,000 per year. We signed it when our SEO traffic was at an all-time high, and it was priced on that traffic. A lot of it came from India and didn't convert, but we paid for it anyway.
Then AI decimated search traffic. Ours dropped like everyone's, and we were stuck in a contract that had us overpaying by something like 400 to 500%. We asked Webflow to work with us on it. They wouldn't.
At that point it was obvious Webflow was no longer a sensible choice. Today it takes almost no time to migrate off Webflow completely and build something better, more intricate and more interesting.
I believe Webflow is a dead company. The product has barely moved in years, and the pricing feels like it's designed for customers who can't leave.
The cost was the trigger, but it wasn't the only reason:
- Animation. We wanted richer motion than Webflow's interactions allow. Now our animations are written in code, so we don't need an animation tool like Rive or Webflow's custom interactions. We just move the pixels ourselves.
- Speed. Every page loaded Webflow's full CSS, jQuery and scripts, whether it needed them or not.
- Ownership. Our images, documents and libraries lived on Webflow's CDN and public CDNs, not on our own domain.
So we decided to leave. But we didn't start with Rezi.ai.
We tested on smaller sites first #
Rezi.ai is our biggest site, so we didn't want it to be the experiment.
We rebuilt two of our smaller brands first: ResumAI and Resumatic. They gave us a safe place to prove the stack, the hosting, the self-hosted files and the page-by-page comparison before any of it touched Rezi.ai.
By the time we started on Rezi.ai, the approach had already shipped twice.
Both sites were designed in Paper and built with Claude Opus 5.5. That workflow deserves its own article, and I'll write one soon. The point here is what they proved.
ResumAI went first. I designed and built it in 4 hours, then shared it on X. The post reached 363K views, with 814 likes and 66 replies.
Resumatic came next, and it was a much bigger test. It took 15 hours and 47 minutes, with a different, more refined workflow, especially on the design side. We tested a lot more than animation:
- A CMS for our content
- Localization. Adding a new language was instant.
- SEO tools built in natively, so the site is always being monitored and updated
- Animation written in code, the kind of motion Webflow's interactions couldn't handle
Once adding a language took no time at all, Webflow's $2,500 a year for one language stopped making any sense.
When I posted Resumatic on X, it reached 1.3M views, with 3.4K likes and 170 replies.
Two launches, real traffic, and nothing broke. Rezi.ai was next.
The rule: no visible change #
Most rebuilds are also redesigns. That's how they go over budget and lose rankings.
We kept them separate. First, move every page exactly as it is. Redesign later, on our own terms.
To make "exactly as it is" real, we wrote a script that compares every page to the Webflow original, element by element. If any page changes in a way we didn't expect, the build fails and nothing ships.
"It looks the same" stopped being an opinion. It became a test.
What we built it with #
- Astro 7. Every page is pre-built HTML, so it's fast by default.
- React 19, only for the parts that need to be interactive.
- Tailwind CSS 4.
- GSAP and Lenis for animation and smooth scrolling, with every animation written in code.
- Netlify hosting, with an edge function for redirects, removed pages and old Webflow URLs.
- Content in Git instead of a CMS: blog posts, help docs, resume examples, cover letters and templates.
- Netlify Forms for the newsletter.
- One Nx monorepo that also holds ResumAI, Resumatic, our job map and our internal growth dashboard.
How the move worked #
- We rebuilt all ~515 pages from Webflow's export and the published site.
- We moved every image, document and JavaScript library off Webflow's CDN and public CDNs onto our own domain.
- We replaced Webflow's JavaScript and jQuery with a small runtime of our own. Each page loads only the parts it needs.
- We turned our free tools (ATS checker, bullet-point generator, resume import) into shared libraries, so every site in the monorepo can use them.
Try the free ATS resume checker →
The hard part: drift #
The biggest challenge wasn't the stack. It was drift.
Drift during the rebuild. A few pixels of spacing here, a slightly different font weight there, a breakpoint that kicks in 20px early. None of it matters on one page. Across 515 pages, it adds up to a site that feels "off" and you can't say why.
Drift after the rebuild. Webflow's visual editor kept the site and the design in one place. Without it, the site can slowly move away from the design one small change at a time.
The comparison script handles the first. Automated checks on every build handle the second.
The results #
Pages are significantly faster:
- Each page carries only the CSS it uses
- Fonts load first
- Images are resized
- Scripts load later
And we fixed SEO problems we couldn't fix on Webflow: crawlable pagination, structured data, a fixed sitemap and proper security headers.
On Google PageSpeed, our performance score went from 38 to 99 on desktop and from 57 to 95 on mobile.
Guardrails that keep it fast #
Every build runs Playwright browser tests plus accessibility, SEO, structured data and security header checks. If something regresses, it doesn't ship.
Webflow made the site easy to change. Now it's hard to break.
What I'd tell anyone thinking about leaving Webflow #
A year ago, a migration like this meant months of work and an agency. We did it in a weekend.
That speed changes how you should work. Rebuilding is so fast now that you should do it more than once. Build it, delete it, build it again. Push until you really see what's possible. The more sites you do, the better each one gets.