I rebuilt my website with AI. 300 prompts, and a 64-to-98 speed jump

Most people think building with AI is one prompt.

You type “make me a website,” something appears, and you ship it. And that is exactly why the internet is filling up with slop.

Because that is not how you get something good. Let me show you what it actually takes.

I just rebuilt my entire website with AI. From scratch. And here is the one number that tells the real story:

300 prompts.

Three hundred. Not one. Each prompt was a moment where I looked at what we’d built and said “yes, keep that” or “no, redo it.” So 300 prompts means 300 rounds of look, react, adjust, and go again.

171 of them happened in a single five-day sprint, concept to live. That’s me and the AI going back and forth, sometimes every few minutes. “Bigger headline.” Look. “No, too big.” Look. “Move the video.” Look. “Now it’s broken.” Fix.

That’s the whole secret, and nobody wants to hear it, because it sounds like work. It is work. It’s just much faster work.

Here’s what came out the other side:

• 65 blog posts moved over from the old site. Every one, images and all.
• 22 case studies, same deal.
• A full site in two languages, English and French. Real pages, not a translate button.
• 36 redirects, so I didn’t torch the Google rankings I’ve spent years building.
• Tens of thousands of lines of code. Thousands of them deleted again, because the first version wasn’t good enough. Neither was the second.

From the first line of code to a live site: under two weeks.

Now let me show you the back-and-forth up close, because one small piece tells the whole story.

The video at the top of my homepage. Sounds simple. It was not.

Launch day, I ran a speed test. Desktop looked great. Mobile came back at 64. Oof. (You want the 90s. 64 means your visitor already left.)

Here’s the thing about fixing page speed. It’s normally my least favorite job on the planet. You go issue by issue, look up what’s even possible, make one change, measure, then do it again. It’s tedious and it eats hours.

This time I just asked Claude to make it faster. And it came back with a clear game plan. It knew exactly what was dragging the site down.

The main culprit was that hero video. On phones it was loading in a way that pushed the whole page to a 12-second paint. Twelve seconds. So we reworked it. A still image loads first, the video kicks in on tap.

(Then the fix broke the video on desktop for ten minutes. So we fixed the fix. Pushed again. The back-and-forth never really stops.)

Then I ran the test again.

Mobile: 98. Desktop: 100. That 12-second load was down to two.

From 64 to 98, and I basically watched it happen.

That’s where AI earns its keep. Not replacing my eye on the design, that took all 300 rounds. Doing the tedious, fiddly work fast, the stuff that used to cost me a whole afternoon.

But I still had to know that 64 wasn’t good enough. The AI would have happily left it there. I’m the one who said no.

And that’s the part I really want you to take with you.

AI did not build this site. AI and I built this site. That difference is everything.

It didn’t know to protect my rankings with those redirects. It didn’t know what “memorable” is supposed to feel like, or where to break a headline so it lands. I knew that. Over a decade of websites and SEO knew that. And I brought it to every single one of those 300 prompts.

So no, AI is not a magic button. It’s the fastest pair of hands you’ve ever worked with. But the hands still need someone with taste, telling them “no, not like that” until it’s right.

That’s the real skill now. Not knowing the most about AI. Knowing what good looks like, and being willing to go back and forth until you get there.

One prompt gets you slop. Three hundred gets you a website you’re proud of.

I’m doing all of this out in the open on YouTube (@webguymike). Come watch me build, break, and fix things in real time.

And if you want a site built like this one, the proper way, with all the back-and-forth, that’s what I do now. Come say hi.

It’s go time.