Insights — Building in public
I built 20 websites in 6 months using AI.
Client sites, internal tools, landing pages. Here's what actually got faster, what didn't move at all, and the tell that gives away an AI-built site.
What actually got built
Over six months I shipped client sites for real estate agents and investors, internal tools for Automize and Kaizen, calculators, landing pages, and a handful of full rebuilds. Every one of them had AI doing the majority of the typing — writing the HTML/CSS, drafting copy, generating the first pass at layout. None of them shipped on the first AI output. That distinction matters more than anything else in this post.
What got faster
The first draft
A site that used to take two or three days to scaffold — page structure, base styling, first-pass copy — now takes a few hours. That's the real unlock. The blank-page problem is gone. I'm never starting from nothing anymore, I'm starting from a rough draft I'm editing.
Variations and iteration
Want to see three different heros for the same page? That used to mean three separate design sessions. Now it's three prompts and a coffee break. The cost of trying an idea dropped close to zero, which changes how many ideas you're willing to try.
Boilerplate that used to eat a day
Privacy policies, terms pages, accessibility statements, FAQ schema, sitemap generation — the stuff that's necessary but nobody enjoys writing. AI handles the first pass on all of it now, and it's usually good enough with light editing.
What didn't get faster
Knowing what to build
AI writes fast. It doesn't tell you whether the site you're building actually solves the client's problem. The strategy layer — what should this page even say, who's it actually talking to, what's the one action we want someone to take — is still 100% a human call, and it's still the part that takes real thinking time.
The last 20%
Every AI-generated site needs a pass for real content (not placeholder text), real photography instead of generic stock, spacing and hierarchy fixes a human eye catches that a model doesn't, and testing on actual devices. That 20% is where most of the remaining time goes, and skipping it is exactly how you end up with a site that looks impressive in a screenshot and falls apart the moment someone actually uses it.
The tell that gives away an AI-built site
I can spot one in about four seconds now, and it's almost never the code quality. It's generic imagery, copy that sounds like it's optimizing for a keyword instead of talking to a person, and layouts that feel like they were designed by committee — competent, correct, and completely forgettable. The fix isn't avoiding AI. It's refusing to ship the first output. Every site here went through at least two full editing passes where I was actively cutting the generic parts and replacing them with something specific to the actual business.
The real lesson
AI didn't make me a faster typist. It made the cost of a first draft close to zero, which means the bottleneck moved entirely to judgment — knowing what's good, what's generic, and what needs to be cut. If you're using AI to build and the output feels underwhelming, you're probably shipping the first draft. Don't. That's the draft AI gives everyone who asks the same question. The edit is where the site becomes yours.