The Growth Framework IG Follow
Mind

Case study: how this website was built with AI tools

This site was built by its owner working with an AI coding assistant. Here is exactly what the AI did, what the human did, what went wrong, and what it cost, as a real example rather than a sales pitch.

· 7 min read · By Waseem Raja

Last checked: October 2026. AI tools change quickly; features and prices may differ when you read this.

The starting point

The Growth Framework began as a single web page with an Instagram account behind it. The owner, who is not a professional developer, wanted proper article pages, better search visibility, an email list and a clearer design. Instead of hiring an agency, he worked with Claude, an AI assistant, through Claude Code, a tool that can read and write code in a project and run commands.

The tools involved

  • GitHub stores the website's code and keeps a history of every change.
  • Vercel hosts the site. Every change pushed to GitHub builds a private preview link first; the live site updates only when a change is approved.
  • Supabase stores email sign-ups in a database, set up so that the website can only add addresses, never read them.
  • Canva was used, through a connector, to generate several of the photos on the site.
  • Connectors (built on a standard called MCP) let the AI use these services directly, with the owner's permission, instead of only giving instructions.

What the AI actually did

  • Wrote the code that turns articles into separate pages, with titles, descriptions and share images for WhatsApp and Facebook.
  • Built the design, the mobile layout, the charts and the light and dark themes.
  • Set up the email database and its security rules, and tested that the public could not read the list.
  • Drafted articles from templates the owner wrote, including sources.
  • Tested every page in an automated browser, checking for broken links, errors and layout problems on phone and desktop sizes.

What the human did

This part is often left out of AI stories, and it matters most.

  • Decided what the site is for, who it serves and what it should not be.
  • Wrote the landing page copy and the article templates and rules.
  • Chose which suggestions to keep and which to change.
  • Reviewed previews and gave the final "go live" approval each time.
  • Owned the accounts, permissions and responsibility for what is published.

What went wrong

  • Free limits ran out. The image tool's AI quota ran out after a few pictures, and another image service's free plan did not allow any generation. The design was adjusted to use code-drawn graphics instead.
  • Some services could not be reached from the AI's working environment, so a workaround had to be built to bring images into the project.
  • Connections expired. Some connectors needed the owner to sign in again before the AI could continue.
  • Facts needed checking. Every health and money claim had to be limited to what reliable sources support, and some details were left out rather than risk being wrong.

What it cost

Vercel and Supabase both offer free plans that are enough for a small site like this. The real costs were the domain name, which is paid yearly, an AI subscription, and the owner's time spent directing and reviewing the work. That time is easy to underestimate.

What this shows, and what it does not

It shows that a non-developer can now build and run a reasonably professional website with AI help, if they are clear about what they want and willing to review the results.

It does not show that AI replaces judgement. The direction, the standards and the checking came from a person. Without them, the same tools could just as easily produce a fast, confident, and wrong website.

Three practical lessons

  1. Write down what you want before you start. Clear rules, like the article templates used here, produce far better results than vague requests.
  2. Always use a preview step. Never let changes go live without seeing them first.
  3. Keep ownership. Use your own accounts for hosting, code and data, so you are never locked in to one tool or helper.

Key idea

AI did most of the typing. The human did the deciding.

Sources

  1. Vercel documentation: Git integration and preview deployments (vercel.com/docs)
  2. Supabase documentation: row level security (supabase.com/docs)
  3. Anthropic documentation: Claude Code and connectors (MCP) (docs.anthropic.com)

Educational content only. This is not medical or financial advice. Check facts with qualified professionals before you act.