Why AI Website Redesigns Without Data Are a Disaster

Section 1: The Temptation

I need to be clear at the start: I'm not against using AI to design and build things. On the contrary, I tell my team to practise with it. Understand the pros, understand the cons, understand the limitations and the opportunities. Find where AI can genuinely improve our design process and drive quality.

But there's a gap between the promise of AI and the reality of deploying a solution that actually works. That gap is where most projects fail.

The Fantasy of Starting Fresh

In the last 12 years working in the recruitment space, I've been involved in thousands of web builds. And there's a pattern I've seen repeatedly.

A company comes to the table with a familiar refrain: "Our website is old. Our sitemap is messy. Our analytics are poor. We have pages nobody uses. We need a complete rebuild."

Before AI, that meant hiring a design agency. The agency would conduct design briefs. They'd audit your sitemap. They'd analyse your Google Analytics. They'd talk to you, understand your business, and make informed decisions about structure and design.

Quite often, they wouldn't strip everything down and start over. They'd enhance and build upon what already worked, improving what had been created over years of iteration.

That was the process. It was slower. It cost more. But it was grounded in evidence.

Now there's a different pitch: AI will build something clean and modern.

And I get why that's attractive. I really do.

The Draw of Speed and Cost

The speed is seductive. Vibe code an entire website in a day. Get to market in 48 hours instead of months. That's a genuine advantage.

We've used this approach ourselves at Sourceflow. Our website is partially vibe-coded because we needed to move fast and market speed mattered more than design perfection. Though I'll note our marketing director would probably argue we still haven't solved the design part.

The cost argument is equally compelling. You can now generate a website design for free or a few pounds that would have cost thousands of pounds from a design agency. It looks good enough to pass the internal test of "oh, this looks professional". And the cost is negligible.

When the cost is essentially zero and the speed is measured in hours, the appeal is obvious.

The Promise: Better Thinking, Fresh Perspective

But there's another layer to the temptation, and it's more insidious.

There's a perception that AI brings fresh thinking unconstrained by how things have always been done. AI doesn't have legacy baggage. It's not attached to old decisions. It sees modern design patterns and best practices that humans might miss.

The pitch is: AI will understand what we need better than we do. Better than our internal teams. Better than any agency we could hire.

There's a promise embedded in this: Claude Design, or Gemini, or whatever AI tool you use, will somehow divine what your brand needs and design it perfectly. As if the AI is a brilliant designer who just happens to know your business better than you do.

Why rebuild on the same dated structure when AI can think bigger? Why be constrained by the past when you can start completely fresh?

The Seductive Framing

The pitch from AI vendors is powerful. Anthropic, OpenAI, and others have created an almost irresistible framing:

"What if we just ask Claude to design your entire website from scratch?"

"Tell it what you do. Tell it your colour scheme. Give it a couple of prompts. And it will build the entire thing for you."

"No constraints. No legacy. Just pure design output as HTML you can put live immediately."

There's a prompt for that. There's literally a prompt that claims to do exactly this.

This framing is so seductive that it's almost impossible not to be drawn in. The promise is clean. The path is simple. The cost is negligible. The speed is extraordinary.

We're being seduced into this without understanding the risks.

The Blind Spot

Here's what people don't understand about how AI actually works:

They assume that because AI can take a URL and some colour schemes and a couple of prompts, and generate something that looks modern, it has somehow understood your business better than you have.

The AI hasn't.

What does it understand about where you've succeeded as a brand? What does it know about your sales points? Your photography? Your office? Your ethos? Your values? The specific design decisions that make you distinctive?

What does it know about the curves of your design? The contour and contrast that has made you what you are?

The answer is: it knows what it can infer from statistical patterns in its training data. It's guessing. It's hallucinating based on how language models work.

But people assume it knows more than it does. They assume it understands their business, their market, their customers, and their competitive positioning.

It doesn't.

The Dangerous Assumptions

Underneath this temptation sit some dangerous assumptions:

Assumption One: The old sitemap was legacy. It wasn't data. It was just cruft accumulated over years.

Assumption Two: The pages that exist now shouldn't actually exist. They're redundant. A fresh start will eliminate them.

Assumption Three: Starting with a blank slate is better than inheriting past mistakes.

Assumption Four: Generic best practices are better than institutional knowledge.

These assumptions are baked into the AI redesign pitch. They're rarely stated explicitly. But they're there.

And they're wrong.

Your sitemap isn't legacy. It's evidence. It's a map of how your users actually navigate your website. It shows which pages matter and which don't. It represents years of learning about what works.

Your existing pages aren't redundant. They exist because they serve a purpose. Yes, some might be unused. But not all of them.

And generic best practices are absolutely not better than institutional knowledge.

Your brand has an ethos. A design language. A way of working. Years of iteration have created something specific to you. That's not legacy. That's your competitive advantage.

The Question Nobody Asks

The seductive pitch promises that AI will think bigger and better than you can.

But it never asks the question that matters: what is the AI not seeing?

What does it not understand about your brand? Your market? Your customers? Your competitive positioning? The reason your design is structured the way it is?

What is it hallucinating? Where is it making assumptions that are wrong?

That's the question that should make you pause.

Because the temptation of starting fresh is powerful. And AI makes that temptation almost irresistible.

But before you give in to it, you need to understand what you're actually throwing away.

Section 2: What You Already Have

So what do you already have that an AI doesn't?

Data. Lots of it. Quantitative data. Qualitative data. Empirical data.

When you fire a prompt into Claude or ChatGPT, that data doesn't come with it. The AI has no access to what you know about your own business. And what you know matters far more than what a general-purpose AI can infer.

Your Sitemap Is Not Legacy—It's Evidence

Sitemaps have been bastardised by SEO agencies for years. People misunderstand what they're actually for. Many carry unnecessary fluff.

But in most cases, your sitemap is one of the most powerful documents you own.

A sitemap is not random. It represents years of learning about user behaviour and website architecture. It's a scaffolding layer that evolves as your website grows. It shows which pages matter because users actually go there, and which pages don't because they don't.

By definition, a sitemap is a web framework that semantically links your website together. If a page doesn't appear in your sitemap, there's a strong chance nobody's actually visiting it—unless traffic is being sent directly from an external source.

For recruitment agencies specifically, your sitemap tells you critical information:

Which pages drive candidate flow versus client flow
Which job listings and content structures matter
The architecture of your SEO and content marketing strategies
What your business is actually trying to achieve
What is valuable to you

You can even use your sitemap for competitive analysis. Look at a competitor's sitemap and you can see where they're investing effort, where they're strong, where you're ahead. You understand their strategy by looking at how they've structured their content.

A messy sitemap doesn't mean your structure is wrong. It means your structure evolved to match reality.

Sitemaps start organised and logical. Then your business changes. A new market opens. A new service launches. A department reorganises. And your sitemap evolves with it. This takes years of learning. It reflects trial and error. It represents what actually works for your business.

That evolution is not a flaw. It's evidence.

An AI redesign that ignores this is throwing away years of learning.

Your Analytics Are a Map of What Works

Open your GA4 account. You have endless information about how people actually use your website.

You have traffic data showing which pages people visit. You have conversion data showing which pages drive business outcomes. You have funnel data showing where users enter, where they go, and where they drop off.

When you combine this with your sitemap, you have something an AI will never have: a precise map of what is working and what isn't.

You know which pages are high-value and which are low-value. This tells you what you should keep, what you should improve, and what you should remove.

More importantly, you know this based on evidence, not guessing.

An AI model has no idea which pages matter to your business. It can't see your analytics. It doesn't know your conversion paths. It doesn't understand your funnel. It's working completely blind to what actually drives your revenue.

When you ask it to "redesign the website", you're asking it to make structural decisions without any understanding of which pages are profitable and which aren't.

That's like asking someone to redesign your office without telling them which rooms are used and which are empty.

Your SEO Equity Is Years of Accumulated Trust

You have SEO data that has taken years to build.

You have rankings for specific keywords. You have backlinks pointing to specific pages. You have internal link structures that have been optimised over time to build trust and guide users through your site.

You have domain authority built over years. Your homepage has more authority than other pages because Google knows it matters. That trust took time to earn.

If you redesign without preserving this structure, you lose ranking equity overnight.

This isn't a minor setback. This is a terminal illness for websites. Recovery can take months or years, if it's possible at all.

An AI redesign that breaks your internal link structure, changes your URL patterns, or reorganises your content will destroy this equity. You might go from ranking on the first page for a critical keyword to page three or four.

The cost of this isn't measured in design hours. It's measured in lost traffic, lost leads, and lost revenue.

Your Design Work Is Validated With Real Users

Outside of hard data, you have something else: years of design decisions tested with real users.

Your team has tried things. Some worked. Some didn't. You've failed, learned, and improved. This happens continuously.

You've optimised performance. You've built accessibility into your design. You've tested usability with real users. You go to events and people tell you "we love your brand". You use that feedback to understand what's working.

Over time, you develop a design language. It's validated. It's distinctive. It works for your market.

An AI doesn't have this. It can't learn from user feedback. It doesn't know what your users respond to. It has no idea what makes your brand distinctive or why certain design choices work for your specific audience.

It will generate something that follows generic best practices. But it won't generate something that's been proven to work in your specific market, with your specific users, for your specific business goals.

Institutional Knowledge Is Real Knowledge

Your team knows what works and what doesn't. Not from abstract theory, but from years of doing it.

Ask any member of your design or marketing team and they'll tell you what failed and what succeeded. They'll tell you why certain pages exist. They'll tell you which design decisions were intentional and which evolved over time.

This is institutional knowledge. It's not documented in an analytics dashboard. But it's real. It's valuable. And it's irreplaceable.

An AI has absolutely none of this.

When you ask an AI to "design your website from scratch", you're asking it to ignore everything your organisation has learned about what works. You're asking it to replace years of accumulated knowledge with statistical patterns from the internet.

What You're Actually Throwing Away

So when you consider asking an AI to redesign your website, understand what you're throwing away:

Years of data about user behaviour (your sitemap)
Years of conversion and performance data (your analytics)
Years of accumulated SEO equity (your rankings and domain authority)
Years of design validation with real users (your brand language)
Years of institutional knowledge (what your team has learned)

An AI doesn't see any of this. It can't access it. It can't learn from it.

It will make decisions in a vacuum. And those decisions, regardless of how modern they look, will be made without understanding what actually works for your business.

That's the cost of the blank-slate redesign.

Understanding that cost is the first step to making a better decision.

Section 3: What Goes Wrong

We're attracted to the shiny new thing. I'm guilty of this too. It's cool to be able to say "I asked AI to redesign my website and it did".

Cool doesn't matter when your rankings disappear, your website tanks, and six years of work go down the drain.

Let's talk about what actually goes wrong.

SEO Collapse

When you redesign your sitemap structure without considering past performance, wins, losses, and learnings, here's what happens:

URLs change. Pages disappear. The internal link structure you've built breaks. The ranking equity you've worked hard to gain vanishes overnight.

Pages that ranked on page one drop to page three or four. With Google's current volatility and the impact of AI overviews, they might disappear from the search results altogether.

Recovery takes months, if not years. In many cases, it's not possible at all. By the time you've tried to re-optimise, a competitor has ranked better with a superior structure and pushed you further down.

You're essentially looking at a complete rebuild. Which is a disaster. It's almost impossible to sell internally without losing all credibility.

For recruitment agencies, there's an additional layer of risk: which keywords are critical to your business? Which keywords can't you afford to lose?

An AI doesn't know this. Not without extensive training and context-specific knowledge. Without feeding it vectorised data about your industry and your market, it's flying blind.

Those critical keywords will disappear from your rankings. And unlike generic keywords, you won't regain them easily.

Yes, if the issue is just meta descriptions or meta titles, you can fix that in a week or two. But if you've combined a complete sitemap restructure with URL changes, internal linking changes, and architectural changes, this isn't a simple fix. It's extremely complicated.

And SEO collapse isn't the only problem.

User Experience Regression

Your user experience has been tailored, tweaked, amended, and improved over years. You might think it's rubbish. You might be right.

But most likely, it works. It probably does the job.

What you don't have is a baseline of how users actually navigate your site. You do have that data. Users have navigated it a certain way, and they found what they were looking for. If they didn't, you would have changed it already.

An AI has no understanding of how users actually move through your website. It will take a shot in the dark, based on generic design patterns, about what users might want to see.

The result: users are now lost and confused. There's no clear internal link structure. There's no journey through your site that you've carefully designed. The information architecture has changed without any user testing, without any humans in the loop.

You've taken 40 pages of custom-designed experiences (custom being used loosely here, since an AI designed them) and deployed them immediately because "AI designed it, so it must be good."

Pages that were easy to find are now buried. The "logical" structure the AI created doesn't match how real humans think.

Here's an analogy: a friend of mine used to own a nightclub. He called it the "toilet test". How hard do people have to try to find the toilet? The intuitive answer is: the path should be obvious. But humans don't think logically. They veer off. They take tangents. They miss obvious signs.

The same applies to websites. Just because you think there's a logical path to information doesn't mean users will take that path.

An AI will create what it thinks is logical. Users will navigate it differently. And you'll have a navigation nightmare.

Immediate Traffic Loss

High-traffic pages are now deprioritised. They might disappear from your navigation altogether.

An AI didn't know which pages were high-traffic. You did. You had that data. You knew which pages were driving business to your business.

But the AI? It didn't see your analytics. So it hid and minimised what actually mattered most.

You see an immediate drop in visitors and leads to your website.

We've seen this with customers who've vibe-coded their websites and then come to us to fix the damage. This is real. It happens. And the impact is immediate.

Accessibility Regression

Our design team is extremely careful about accessibility. We take WCAG compliance seriously.

If you built accessibility into your old website, I will almost guarantee that generic AI output will not have the same level of compliance. Even if you prompt it. Even if you ask specifically for accessible design.

You lose WCAG compliance. You lose users with disabilities. And you lose legal protection.

That last one is critical and rarely discussed.

If a user with a disability sues your business for not meeting accessibility standards, "I used AI, so it must be Claude's fault" is not a defence that will stand in court.

But "we have a design team, we use testing tools, and we can evidence the work we've done" carries significantly more weight. You can show your working. You can prove diligence.

It's like showing your working on a GCSE maths exam. The evidence matters.

Conversion Rate Collapse

You've spent years optimising your job application flow. You know what questions matter. You know what sequence works. You know what converts candidates.

An AI will replace your optimised flow with "best practices". Generic guesses about what users might want to see next.

The AI doesn't know what people actually want to see next. It's not human. It doesn't have your data.

Your conversion rate drops. Revenue is immediately impacted.

Recruitment-Specific Failures

Here's what an AI fundamentally doesn't understand: the recruitment industry.

It doesn't understand what breaks when you reorganise candidate flow. It doesn't understand the impact of changing how clients move through your site. It doesn't understand your business model.

It doesn't understand job URL structures. It doesn't understand how your CRM maps jobs into your systems. It doesn't understand that you've just started recruiting in a small township outside Northampton where a new factory opened, and you need an SEO presence for that specific market.

These are recruitment-specific challenges that require domain knowledge. An AI will still output something. It will create something that looks plausible. It will do things you want to hear. It will make you happy.

But it won't be right. And you'll only discover that after you've deployed it.

Best Case vs. Worst Case

The best case scenario: your new website looks plausible. You start to rebuild from scratch. You go through another redesign. It's frustrating and expensive, but you recover.

The worst case: you tank a website that's been performing well over years. You lose rankings. You lose traffic. You lose revenue. Recovery takes months or years.

And in between, you're explaining to your board why you decided to throw away a working website for something an AI generated.

That's the conversation you don't want to have.

Section 4: The Right Framework

So how do you actually use AI properly? Not as a replacement for thinking, but as a tool for optimisation.

The answer starts with a simple principle: data first, AI second.

Start With Your Data

Before you ask an AI anything, ask your data.

What do you want to know about what you have? What can you use this information for to make version 2, 3, 4, or 5 of your website better?

Ask your sitemap: What does it tell you about your structure? Which sections matter? Where are your content clusters? AI can actually help here—it can analyse your sitemap structure quickly, identify groupings, build themes and links. That's a legitimate use of AI: rapid analysis of existing data.

Ask your analytics: What value is your website creating? What pages drive conversions? Where do users behave in ways that surprise you? What's working and what isn't? This data tells a story about your business that AI doesn't know.

Ask your SEO data: What digital equity has your brand built over time? What does Google want to promote you for? What pools of people does it think should see your website? What's working according to search engines, and what's broken (indexing issues, ranking drops, etc.)? This tells you what you must preserve in version two.

Ask your design history: What did your website look like last time? What did you include that you later changed? What did you leave out that might have improved things? Your design history is a record of learning. Don't ignore it.

When you introduce a data-first approach, you start to build a rigorous map of what has and hasn't worked for your brand. What's currently working. What the sentiment and emotion behind your brand actually is.

You do this before you feed anything to an AI. Before you start building frameworks. Before you prompt any models.

This is foundational. Everything else builds on this.

AI as Constraint Solver, Not Replacement

Here's the critical shift: don't ask AI to redesign from scratch.

If you do, you'll get the same website as everyone else. Sure, your prompt might be slightly different. But everyone's prompts are slightly different. And you'll all end up in the same generic space.

You lose the moat around your business. You lose your brand differentiation. Your brand is built on data—on years of learning what works for your specific market, your specific users, your specific business model. Throw that away and you're just another website.

Instead, ask AI to optimise within constraints.

Here's the difference:

Wrong question: "Design my entire website from scratch."

Right question: "Given that these pages must exist, this user journey must work, and this SEO must be preserved—how do we improve the design?"

That's a constraint-based question. You're giving AI the boundaries. You're saying: here's what we know works, now help us make it better.

This is fundamentally different. You're not asking AI to think bigger than you. You're asking it to optimise within what you already know.

And that's where AI actually adds value.

Work Modularly, Not Monolithically

Instead of redesigning 40 pages at once, redesign page by page. Test changes incrementally. Validate each change against your data.

This is slower than throwing everything at AI and hoping it works. But it's faster than recovering from a complete redesign disaster.

Be Skeptical of AI's Recommendations

Here's something important: AI will find problems with almost anything you put in front of it.

Someone on LinkedIn recently posted about putting their design into an AI model. The AI gave them a list of UX problems. They took all the recommendations. They asked the AI to make those changes. Then they fed the updated design back to the AI.

And the AI said the changes were wrong.

AI is cyclic. It doesn't always tell you the right answer. It tells you what it thinks you want to hear. It's not confrontational by nature.

So when AI recommends changes, ask it to evidence why. Don't just accept its recommendations. Challenge it. Ask it to explain its reasoning. Make it justify its suggestions with data, not intuition.

Because AI's intuition isn't always reliable.

Left Brain vs. Right Brain

AI is extremely good at left-brain thinking: logic, structure, optimization, pattern recognition.

Right-brain thinking—creativity, intuition, emotional resonance, brand feeling—is still far behind human capability.

This is why you can't replace humans with AI in design. You need both. AI handles the logic. Humans handle the creative vision.

The Platform Question

At Sourceflow, our approach is this: vibe code quickly to test concepts. Then deploy through a proper platform.

Vibe coding gets you to market fast. But it doesn't get you to market right.

A platform like Sourceflow handles what AI doesn't: accessibility, performance, scalability, security, monitoring. All the things that transform a prototype into production-ready infrastructure.

You could just deploy vibe-coded HTML to a server. Sure. But you'd be abandoning accessibility. You'd have no performance monitoring. You'd have no security layer. You'd have no way to scale.

Why would you do that?

Instead: vibe code something at small scale, test the concept, then integrate it into a proper platform. The platform applies your accessibility standards. It handles your performance requirements. It manages your security. It monitors what's actually happening with users.

We did this with Sourceflow Deployed. You vibe code quickly. You pop it into our system. You have a website live with proper infrastructure backing it.

Is it faster than custom building from scratch with a proper design team? Honestly, I'm not sure it is. But it's a different approach with different trade-offs.

The point is: don't just vibe code and deploy. Run it through a proper platform.

Questions You Must Answer

Before you ask AI to do anything, answer these questions with your data:

Which pages are non-negotiable? High traffic. High value. High conversion. These pages must exist and must work as they currently do.

Which user journeys must be preserved? Your candidates' journey. Your recruiters' journey. Your clients' journey. These aren't random. They exist because they work.

Which SEO rankings are critical? Which keywords drive business? Which rankings can't you afford to lose? These are your priority pages.

What's your accessibility baseline? If you've built accessibility into your site, you need to maintain it. Not regress from it.

What design language has been validated? What do your users respond to? What feels right for your brand? What's been tested with real people?

These questions are the backbone of any AI prompting you do. If you're going to use AI, these are your constraints.

And honestly, there are probably 50, 60, or 100 more questions you could ask if you dug deeper.

This interrogation process is absolutely no different than you would do going into a redesign with human designers. It's just more rigorous. And it's essential.

Because if you don't know the answers to these questions, you're not ready to redesign anything—whether you use AI or not.

The Core Principle

Data informs constraints. AI optimises within those constraints.

Not the other way around.

The moment you let AI set the constraints is the moment you lose.

Section 5: Implementation & When Blank Slate Works

How to Actually Do This

If it were me working on this, here's the step-by-step process I'd follow:

Step One: Data Analysis Answer the core questions from the previous section. What pages matter? What journeys work? What SEO equity exists? What's your accessibility baseline? What design language has been validated?

Step Two: AI-Assisted Analysis Use AI to help analyse your sitemap and content. Ask it to pull out themes, content blocks, structural patterns. This is a legitimate use of AI—rapid analysis of your existing data.

Step Three: Validate, Don't Just Accept Create constraints from that analysis. Then validate them with humans. Don't just accept what AI tells you. Make sure the findings are real, not just what you want to hear.

Step Four: Optimise Within Constraints Ask AI to optimise page by page or section by section. Given these constraints, what's the best logical thing to display? What could we improve?

Step Five: Test Incrementally Don't deploy everything at once. Test changes incrementally. See if they work. Gather data.

Step Six: Platform Hardening Run it through Sourceflow (or a similar platform) for production hardening. Make sure it actually works in a real environment. This is critical. Vibe code is not production-ready. A platform makes it production-ready.

Step Seven: Continuous Improvement Keep checking performance, user experience, and behaviour. Continue the cycle page by page.

This is more work than asking AI to redesign from scratch. But it's also infinitely less risky.

The Speed vs. Risk Trade-Off

Is this faster or slower than a blank-slate approach?

Honestly, I'm not sure.

You can get to market quickly with AI. I get the attraction. I feel it too. As a busy marketer or CEO, speed is seductive.

The issue is: are you getting to market quickly at the start, then slowing down massively as you discover everything's broken?

That's the real timeline.

Data interrogation takes effort. Significant effort. But it's worthwhile effort. It's the effort that prevents websites from tanking.

There's a trade-off here between risk and speed.

If you don't mind risk, speed is great. You get there fast. Something breaks, you fix it later.

If you're worried about risk, speed is not your friend. High-speed approaches come with large misgivings.

You need to decide which you value more.

When Blank Slate Actually Works

I've spent a lot of this article talking about when blank-slate redesigns don't work. But they do work in specific circumstances.

Blank slate works when:

A new product line with no historical data exists. A new website for a new market with no historical data. Something brand new to the world with no existing performance data.

In these cases, vibe coding is a legitimate first step. You have no preconceived ideas. You have no data to constrain you. You're building from zero.

We did this with Sourceflow Deployed. We needed to get to market quickly with something new. We vibe-coded it. We launched it. We're building data now.

Vibe coding is also fine for experimental areas. Throwaway prototypes. Things you're testing to see if they work. I encourage this. Fail fast with low-risk experiments.

But here's the caveat: if you vibe code something brand new and it doesn't work, we're not responsible for that failure. AI did the work. We provided a secure platform and support. But the outcome is on the AI, not on us.

And there's another important caveat: if you vibe code something brand new, it will be generic. You won't get differentiation. You'll get something that looks like everything else.

If you're okay with that, and you want to get to market quickly, vibe something together and put it on a Sourceflow platform. Let's see how you get on. We're game for it.

But if you want something distinctive? Something that actually leverages your competitive advantage? Something that your users won't find anywhere else?

Don't use blank-slate AI redesign.

Blank slate does NOT work when:

Existing customer-facing experiences are involved. Anything with existing traffic. Anything with existing rankings or conversions.

In these cases, you follow the framework above. Data-first. Constraints. Optimization. Testing.

If there's something brand new to the world with no existing performance data, absolutely—fire it out and see what happens. Start your journey of building data. But for anything with existing performance? Use the framework.

The Role of the Platform

At Sourceflow, our approach is this: vibe code the HTML and CSS. Import it into our platform. We do some configuration on our end to make it work.

We're doing this now with customers. We've done it with ourselves. The advantage is real: you get the security of a proper platform plus the flexibility to add pages, manage content, and iterate.

AI won't enable you to do this quickly. But a platform will.

A platform also handles all the architectural issues and headaches that have consumed my last 25 years of working with web infrastructure. Infrastructure management. Component handling. Page building. Daily operational stuff that you want to automate, not manually maintain.

The vibe code ends where the platform begins.

This applies to forms too. When you vibe code a form, the platform takes over. Data capture. Security. Compliance. All handled by the platform, not by your manual processes.

The platform does things that AI simply doesn't do: hosting infrastructure, component management, page building, day-to-day operational management.

Does this mean you go significantly slower? No. It means you get a much more robust model to launch from and build on.

When This Framework Applies

So when should you use this data-first, constraint-based approach?

Use it for:

New website projects (anything you're building from scratch)
Redesigns and refreshes of existing websites
Anything with existing traffic, rankings, or conversions
Any customer-facing experience that matters to your business

You can skip it for:

Microsites with no existing audience
New business lines or territories with no historical data
Regulatory compliance projects where you're launching something completely new
Throwaway experiments and prototypes

For everything else? Do the work. Interrogate your data. Create your constraints. Optimise within them.

The Core Message

I'm not against vibe coding. I'm not against AI building things. I use it. I encourage my team to use it.

But vibe coding an entirely new version of an existing website is terrible and should be avoided at almost all costs.

Unless your existing website has gone through some kind of existential crisis and decided to blow itself up, do not start from scratch.

Use the framework. Do the work. Build constraints from your data. Optimise within them. Test incrementally. Run through a proper platform.

It's slower at the start. But it's faster at the finish.

And it's infinitely less risky.

Photo by Ales Nesetril on Unsplash