Writing

AI-First Web Development and the Things You Don’t Know to Ask

A man works at a laptop while a humanoid robot beside him points at a floating code editor. Around them float a finished website layout and a phone view of the same site, with icons for settings, site structure, security, and page speed.

I went looking at website trends for the year, mostly out of curiosity, and the first thing on every list was AI development. No surprise there.

What I keep running into is the assumption that comes with it. People like me are about to be out of a job. Anyone can describe a website now and get one back, so what exactly is the web guy for?

I want to be honest about the first part before I argue with the second. The way I work has shifted dramatically over the past year. Not a little. Dramatically. Anyone in this field telling you their process is unchanged is either not paying attention or not being straight with you.

But here is the thing I did not expect. The way I use AI and the way most people use it are almost complete opposites.

People Use It for the Part They Can See

If you do not build websites for a living, a website is what it looks like. That is not a criticism. It is just how anybody outside this work experiences a site. Layout, colors, images, the words on the page. That is the website.

So when someone sits down with an AI tool, that is what they ask for. They describe what they want it to look like, they iterate on the design, and when it looks the way they pictured it, they are done. They imagined it and it exists. It feels like the whole job just got solved.

People are getting ambitious with it too. I am seeing folks attempt genuinely complex sites, the kind of thing that would have been a real project a couple of years ago, on the theory that if they can describe it, they can have it.

Sometimes that works out fine. The part that concerns me is everything that never came up in the conversation.

I Use It for the Part Nobody Sees

I have been building more coded sites lately, and AI has become a real part of how I do it. Not for the design. I know what I want a site to look like and I do not need help getting there.

What I use it for is the invisible layer underneath.

Security is the big one. Headers, form handling, the specific ways a contact form can be turned into a spam relay if nobody thought about it. This is the stuff that never shows up on a screenshot and never comes up when someone is picking fonts.

Structured data is another. Telling search engines what this business is, where it operates, and what it does, in a language they actually parse. A site can look perfect and still be effectively mute to Google.

SEO scaffolding. Titles, metadata, heading structure, internal linking, the sitemap. All the plumbing that determines whether a site is findable at all.

Then there are the quirks. iPhone rendering behavior alone has cost me more hours over the years than I care to add up. Safari on iOS does things its own way, and a site that looks flawless on a desktop can be subtly broken on a phone in ways you will never notice unless you know to check. AI is genuinely useful for catching that class of problem.

None of that is visible. That is exactly why it gets skipped.

The Content Version of the Same Problem

I have already watched a smaller version of this play out with writing.

People have brought me AI-generated content, one prompt in, output out, and it needed a lot of work. It was not unusable. It looked like a blog post. Paragraphs in the right places, headers where headers go, the general shape of the thing was correct.

It just was not right. Facts that needed checking. Claims a regulated business genuinely cannot make. A voice that belonged to nobody. It had the form of finished work without any of the judgment that gets you there.

Same pattern as a website. Lower stakes, easier to see.

The Bill Comes Later

What makes this tricky is that the problems do not surface on day one.

A site with no structured data, no proper metadata, and a sitemap nobody submitted still loads fine. It looks finished. The owner is happy. The trouble shows up months later when the phone is not ringing and nobody can explain why, because the site was never really built to be found in the first place.

Same with a form that submits into nothing. It looks like it worked. There is a thank you message and everything. Nobody finds out until a customer mentions they reached out weeks ago and never heard back.

That is the actual cost, and it is a delayed one. Which is exactly why it is so easy to skip.

Generating and Knowing Are Different Things

I am not going to tell you AI cannot build a website. It can, and it is getting better at it constantly.

What it cannot do is want the right things on your behalf. It answers the question you asked. If you did not know to ask about security headers or structured data or how the site behaves on an iPhone, it is not going to raise its hand and bring them up.

That is the whole gap. It is not about whether the thing can get built. It is about knowing what to ask for.

When something goes wrong on a site you generated but never understood, you are stuck. You cannot fix what you cannot read.

AI made it easy to produce a website. It did not make it easy to know what a website needs. Those were always two different problems, and the second one determines whether the thing actually works.

Call 516-880-4201 or get in touch if you want someone to look at what you have and tell you honestly what is underneath it.

Talk about your site More writing →

Related writing