We spent about $500 a year on Webflow for a site we barely changed. The front end looked decent enough, but every time we wanted to update a paragraph or add a section, we had to log in, click through a visual editor and wonder why we were paying for drag-and-drop boxes when all we needed was text on a page. So we rebuilt it. Not with another CMS—just static files, generated by an AI agent and hosted on Cloudflare's free tier. The whole thing takes three accounts and about thirty minutes.

This is not a tutorial about leaving Webflow. It is about something more useful: connecting your site to an AI agent so you can prompt changes instead of clicking around an editor. The Supernodes approach treats your website as a codebase that any AI assistant can read and modify, which changes how fast you can iterate.

As of June 2026, Cloudflare Pages serves over 15 million sites on its free plan, and GitHub hosts more than 100 million repositories. Both are already part of most engineering teams' workflows. The missing piece is the AI agent that ties them together.

The real cost of visual editors

Webflow, Wix and Squarespace sell you convenience. You pay a monthly fee so you can drag elements around a canvas instead of writing HTML. That trade-off makes sense when you need a visual designer who does not code. But if your team ships text updates, new landing pages, and campaign-specific content, the visual editor becomes overhead.

Every change means logging in, finding the right page, waiting for the editor to load, clicking the right element, and updating the text. Then you publish, wait for the build, and check it live. With a static site and an AI agent, you tell the agent what to change and it edits the file directly. Then you push to GitHub and Cloudflare deploys it. That is it.

Cloudflare's network spans 330 cities and serves static assets at the edge with zero cold starts. A static HTML file on their CDN loads measurably faster than a CMS-managed page that has to query a database and render templates on each request. The performance difference alone is worth the switch for most content-driven sites.

Three accounts, thirty minutes

Here is exactly what you need and how to set it up. You can use any AI agent that saves files locally — Claude, ChatGPT, Codex, Cursor, or Copilot all work.

Step 1: Feed your current site to an AI agent and let it rebuild locally

Open your AI agent of choice — Claude Code, Cursor, ChatGPT, or Codex. The key requirement is that the agent can create and save files on your machine. Claude Code and Cursor both support this natively. ChatGPT requires the Codex interpreter. Codex works out of the box with file operations.

Give the agent your current website URL and this prompt:

"Visit this URL: [your website URL]. Analyse the page structure, every section, the images used, the copy and the tone of voice. Then recreate the entire site as a set of static HTML, CSS and image files on my desktop. Keep the design, layout, images, copy and tone identical. Save all files in a folder called my-site."

A capable model like Claude Opus 4.6 or GPT-5.5 handles this well in a single pass. If the result is not pixel-perfect on the first attempt, give the agent feedback: "The header is too tall, make it match the original spacing" or "The font sizes look different, use the same sizes as the original." Two to three rounds of feedback usually produce a near-identical replica.

Once you are happy with the output, you should have a folder on your desktop with an index.html file, a css/ folder, an assets/ or images/ folder, and any additional pages you asked the agent to recreate. Verify that opening the local index.html in a browser looks correct before moving to step 2.

Step 2: Initialise a Git repository and push to GitHub

Open a terminal in your site folder and run these commands:

cd ~/Desktop/my-site
git init
git add -A
git commit -m "Initial site from AI rebuild"

Now create a new repository on GitHub.com. Do not add a README, .gitignore or license — you want an empty repo. Copy the remote URL it gives you (it will look like https://github.com/yourusername/your-repo.git). Then run:

git remote add origin https://github.com/yourusername/your-repo.git
git branch -M main
git push -u origin main

If your agent supports GitHub authentication (Claude Code and Cursor both do), you can skip the terminal and ask the agent to do this for you: "Create a GitHub repository called my-site and push all the files in this folder to it." The agent handles authentication and remote setup automatically. GitHub's free tier includes unlimited private repositories for teams of up to 15 people, so there is no cost for this step.

Step 3: Deploy with Cloudflare Pages

Log in to your Cloudflare dashboard at dash.cloudflare.com. Navigate to Workers & Pages and click Create. Select Pages, then "Connect to Git." Authorise Cloudflare to access your GitHub account and select the repository you just created.

Under build settings, leave the framework preset as "None" (your site is plain HTML) and set the build output directory to the root (/). Click Deploy. Cloudflare pulls the files from GitHub, pushes them to their global CDN across 330 cities, and gives you a .pages.dev URL within about 30 seconds.

To use your own domain, go to the Pages project's Custom Domains tab, enter your domain, and follow Cloudflare's DNS instructions. Cloudflare handles SSL certificate provisioning automatically — no manual certificate setup required. The whole process from git push to live domain takes under two minutes.

From this point forward, every time you push a change to the GitHub repository, Cloudflare Pages automatically rebuilds and redeploys the site. There is no manual deploy step. No build server to maintain. No subscription fee.

Once deployed, your site is fully promptable. Change the hero section by telling the agent. Add a new page by describing it. Rebuild the whole site for a campaign by explaining what you need. The agent edits the files, you push to GitHub, and Cloudflare updates the live site. A typical update takes about a minute from prompt to production.

Where the pattern compounds

For a personal site this is worth doing for the speed alone. But the setup becomes genuinely powerful when you scale it.

E-commerce teams can attach a spreadsheet with their product catalogue and ask the agent to generate one page per product. Pricing updates become a CSV upload instead of a manual edit across hundreds of product pages. Stock changes, seasonal promotions, flash sales — all batch-generated from structured data rather than hand-edited through a CMS interface.

Sales and marketing teams can generate personalised landing pages for every person they reach out to. Company name, industry, role — pulled from a spreadsheet and rendered into a unique page on your domain. That level of personalisation is impractical through any visual editor. With an AI agent it is a prompt and a data file.

Content teams can batch-generate pages from an editorial calendar, review them in a single pass, and publish the whole set in minutes. No logging into a CMS, no clicking through templates, no waiting for the page builder to load.

This is not a workflow builder. It is text files, a deployment pipeline, and an AI agent that connects them. Any modern LLM can do it.

What you lose and what you gain

Static sites do not have a visual editor. If your team includes non-technical members who rely on drag-and-drop layouts, this approach requires an extra step: a preview environment where they can review changes before they go live. You can set that up with a staging branch on GitHub and Cloudflare's branch deployment feature.

You also give up built-in CMS features like user permissions, scheduled publishing, and draft workflows. These can all be handled with lightweight tooling — GitHub for version control and access management, Cloudflare for branch-based previews — but it is worth knowing you are trading convenience for speed.

What you get instead: zero hosting cost, faster page loads, full control of your files, and the ability to change anything on your site with a single prompt. For teams that iterate fast — launching landing pages, running campaigns, testing copy — that trade-off pays for itself in the first week.