I’ve been using AI longer than most people.
Not just because I’m usually early to trends, but because I needed it.
In 2017 I suffered a concussion, and for years afterward I dealt with issues that were hard to explain unless you’ve lived it: focus, concentration, vocabulary recall, and the ability to take what I was thinking and convert it into coherent writing. It wasn’t just “brain fog.” It was like losing access to parts of my own mind.
By late 2020 / early 2021, I started using early AI tools that predated ChatGPT and the mainstream wave of OpenAI adoption. The real value wasn’t that AI had answers. It was that it helped me organize my thoughts into something structured and readable again. It didn’t replace thinking — it helped me organize it.
So when ChatGPT exploded in 2022 and 2023, I wasn’t surprised. I was curious. And I started asking the same question I’ve asked about every major shift in technology:
What will this do to the systems we’ve been relying on?
For me, one of those systems is web development.
At SOLUENCY, we began as a web hosting company. Over time we expanded into DevOps as a Service, then web and mobile app development. So I’ve spent years on both sides of this world — infrastructure and application layers, production systems and customer demands. And like a lot of people who’ve built websites for a long time, I’ve had complicated feelings about how websites are created.
Especially WordPress.
I don’t like WordPress for static websites. I understand why it dominates. I understand why it’s convenient. But I don’t like what it turns into in practice: performance problems, security problems, plugin dependency, and a general sense that most sites are held together by a stack of decisions nobody remembers making.
I prefer simpler approaches when the job allows it: React, Bootstrap, static HTML — things that can be served cleanly without an entire ecosystem attached.
And for a while, I assumed that AI would make this easier.
I was wrong.
The first mistake I made with AI and websites
In late 2024 and most of 2025, I dabbled in using AI to create websites the way most people do:
- using purpose-built “AI website builders”
- using ChatGPT directly
- using Grok directly
- asking for “a website for X business”
- trying to iterate the design in a single thread
And I almost never liked the results.
They weren’t terrible in the way a broken website is terrible. They were worse.
They were generic.
They looked like a thousand other websites. They didn’t reflect the soul of the project. They didn’t reflect the tone I wanted. They didn’t reflect judgment. And trying to steer them toward something specific was exhausting, because I was spending all my energy correcting the AI’s assumptions.
At some point I realized something:
I wasn’t failing because AI couldn’t write code.
I was failing because I was trying to make AI build a website — when what I actually needed was for AI to help me design the system that builds the website.
That shift changed everything.
The breakthrough: using AI to write the prompt
A few weeks ago, I tried a new approach.
Instead of opening ChatGPT and saying “build me a website,” I went to Grok and said something closer to:
I need you to write me a detailed prompt I can paste into another AI to generate a Bootstrap website for Restore Creation. Here’s what I’m hosting on, here’s the purpose, here are details — and I want the output as files I can upload and run.
I gave it a jumble of information: what Restore Creation is, what kind of tone I wanted, what kind of structure I wanted, what tech stack I wanted. I didn’t try to make it pretty. I just poured everything in.
And Grok returned a prompt that was more than two pages long.
It was extremely detailed. It included the kinds of specifications I would normally write as a design doc: layout guidelines, content organization, formatting expectations, files, naming, structure, and code requirements. It was the kind of prompt I never would have written that well — and definitely not that quickly.
I copied that prompt into a Word document, tweaked it lightly, then pasted it into ChatGPT.
And what ChatGPT produced was impressive.
Not because it was magical.
Because it was usable.
It took what would have been weeks of effort — not just coding, but designing, structuring, rewriting, and iterating — and it condensed that into minutes.
For Restore Creation, it generated a clean one-page Bootstrap site that felt aligned with the project. I could drop the files on my server and it worked.
That was the first time AI website generation felt “real” to me.
Not because AI suddenly got better.
But because I finally had the right system.
Testing the system: VITAL
After discovering this, I needed to know if it was repeatable.
So I tested it again — this time for a far more complex use case: the VITAL marketing website.
The VITAL site had been on our roadmap since at least May 2025.
We tried several approaches:
- trying to generate landing pages with AI
- considering a purchased theme to modify
- trying to find a theme that didn’t feel like a square peg in a round hole
- having a digital agency customer attempt a UI design in Figma we could translate and build
Nothing gained traction. Not because any of these were impossible, but because they required sustained time and attention and like most small companies, we were constantly pulled back into paid client work and operational fires.
Eventually, I decided to apply the same system I used for Restore Creation.
But this time, I took it much more seriously.
For VITAL, I always envisioned something different from typical marketing sites.
Most websites have a CTA like “Book a demo.”
I didn’t want that.
I wanted people to test-drive the technology.
So I went back to Grok — but instead of feeding it a jumble of thoughts, I fed it a PDF that explained what VITAL is: the use cases, the markets, the functionality. Then I told it:
- I want a Bootstrap multi-page site
- I want a three-point call to action: Call / Text / Live Chat
- and I want those channels to route to AI sales agents
- those agents were programmed to role-play and simulate how the system would handle a real small business inquiry
- answer objections
- and attempt to book the appointment that would ultimately close the sale
Grok returned another prompt — again over two pages long.
I lightly edited it.
Then I pasted it into ChatGPT.
And it blew my mind.
Not because it was flawless.
But because it got me about 98% of the way there.
ChatGPT coded, live in front of me, what would have taken months of full-time effort to design, build, and iterate. It did it in under ten minutes. I uploaded the files to the server, and it worked.
And what it produced wasn’t a generic template.
It was my vision.
The “test drive VITAL” idea was implemented the way I imagined it. The structure was there. The messaging was close. The site felt like it belonged to the product. I still had to tweak and revise parts — but the hard part wasn’t “how do I build this” anymore.
The hard part was simply: what do I want?
That’s a different kind of work.
What changed (and why this matters)
This experience clarified something for me.
AI is not the replacement for web developers.
AI is the replacement for the blank page.
Most people treat AI like a magic box that outputs websites.
But that’s not what it is.
AI is a compression engine for thinking.
It collapses:
- brainstorming
- architecture
- implementation
- editing
- translation (idea → code)
…into one workflow.
But that only works if you treat the workflow like a system.
If you treat AI casually, it produces casual results.
If you treat it like engineering, it can produce engineering-level output.
The constraint becomes less about ability and more about judgment.
The system (a practical tutorial)
Here’s the system I’m using right now. It’s still evolving, but it works — and it’s repeatable.
1) Decide what you are actually building
Before you involve AI, get clear about the type of site:
- one-page landing page?
- multi-page marketing site?
- blog?
- membership site?
- web application?
This matters because different systems require different structures.
2) Pick the right foundation (I recommend Bootstrap for most static sites)
For simple marketing sites, I prefer Bootstrap because:
- it’s fast
- it’s responsive by default
- it doesn’t require a build system
- it can be deployed as static files
- it’s easy to modify later
- you can host it anywhere
If you’re building a web application, that’s different — but for “website websites,” Bootstrap is an ideal system.
3) Use AI to write the prompt (don’t write it yourself)
This is the key.
Instead of trying to prompt the site into existence, ask the AI to write the prompt spec.
In other words: make AI produce the requirements doc.
I used Grok for this step, but it doesn’t have to be Grok. The important thing is:
- prompt should be long
- prompt should be structured
- prompt should include constraints
- prompt should be explicit about deliverables (file outputs)
- prompt should include tone / voice requirements
You want something that reads like instructions to an engineer, not a request to a chatbot.
4) Edit the prompt in a Word doc (or any editor)
Don’t paste it directly.
Put it into a document and clean it up:
- remove contradictions
- fix unclear phrases
- add any missing requirements
- tighten language
You’re doing “requirements engineering” here.
5) Paste the prompt into a NEW AI chat
This matters more than people realize.
Don’t paste it into an AI thread that already has noise.
Start a new conversation and paste the prompt cleanly so the AI doesn’t inherit assumptions from earlier context.
6) Require output as real files
Be explicit:
index.htmlabout.htmlcontact.html(or other pages)- a
styles.css - clear directory structure
Then deploy it and test it in a browser.
7) Revise like an editor, not a coder
This last part is important:
Most people try to revise by tinkering.
I revise by specifying:
- what I like
- what feels off
- what needs to be moved
- what needs to be removed
- what needs to be simpler
Then I have the AI implement the edit.
You’re not “coding.” You’re directing.
A note about prompts (what I haven’t tested yet)
I should mention one thing I haven’t tried:
I haven’t tested creating the long “prompt spec” inside ChatGPT itself.
It might work. It probably would.
But right now, I like separating the steps:
- One AI generates the requirements prompt
- Another AI executes it
Because it reduces confusion and prevents the AI from half-building the site while trying to spec it.
If you try this approach, I strongly recommend:
- generate the prompt in one chat window
- paste it into a completely separate chat window for execution
That separation is part of the system.
Where this goes from here
I’m sure this system will change.
AI will get better. Browsers will change. The tools will shift. And some of what I’m doing now will look obvious in hindsight.
But for the first time, I feel like I have a reliable way to build websites quickly without sacrificing standards.
And that matters, because it changes how fast ideas can become real.
Not everything needs to be a massive software project.
Some things just need to exist:
- clearly
- quickly
- without bloat
- with the right tone
- with room to grow
This is one of the best examples I’ve seen of a system doing what it was designed to do — once it’s applied correctly.