Articles Article
The Content Debt Nobody Budgeted For.
Vibe coding made websites cheap to build. Turns out it's made them expensive to own. And the bill is landing on clients.
6 min read

I've been having the same conversation with brand folk for months now. It goes something like this.
A microsite knocked out in a weekend by a freelancer and an AI agent. A campaign page the in-house team prompted into life. A product hub from another agency again. All shipped fast. All looked great on launch day. High fives all round.
Then someone in marketing asks a simple question. "How do I change the headline?"
S i l e n c e.
That moment, playing out across thousands of brands, is the client-side cost of non-standardised content management. IMO it's the hangover of the vibe-coding boom. And it's not a project problem any more. It's an industry one.
How we got here
In under two years vibe coding went from a cheeky phrase to the way a lot of web work gets made. Karpathy coined it in February 2025. Collins made it Word of the Year. A quarter of one Y Combinator cohort shipped codebases that were 95% or more AI-generated (Cloud Security Alliance).
I get the appeal. Brief goes in, working site comes out, cost falls through the floor. Freelancers, in-house teams, production shops and agencies all grabbed the same tools at the same time. Fair enough.
What nobody coordinated was the after. Every builder made their own calls on where content lives, how it's edited, who can publish and where it's hosted. Half the time those calls were made by the model, not the person. Multiply that across every campaign, market and brand a client runs and you get a digital estate with no common shape. Kinda like a house where every room was built by a different builder.
What the client inherits
Here's the thing. The client doesn't inherit a website. They inherit a bunch of one-offs, each with its own logic. This is what I'm seeing inside a typical mid-sized brand after a year of fast AI builds.
Property | Built by | Where the content lives | How marketing edits it | Who can fix it |
Main brand site | Lead agency | Hard-coded in components | Raises a ticket | The original developer |
Product launch microsite | Freelancer with an AI agent | JSON files in a repo | Can't | The freelancer, if reachable |
Campaign landing pages | In-house team | A no-code builder | Drag and drop, off-brand | Whoever has the login |
Regional sites | Second agency | A headless CMS, custom schema | A form nobody was trained on | The second agency |
Events hub | Production shop | Markdown plus a database | Asks someone to re-prompt | Unclear |
None of these is wrong on its own. Each one made sense for that project, that budget, that deadline. The trouble is the sum. Five properties. Five content models. Five editing experiences. Five hosting setups. Five sets of people holding the keys.
That's non-standardised content management. Not a bad CMS. No shared CMS at all.
The industry-wide effect on clients
One brand with a messy estate is a headache. Thousands of them is a structural cost on the whole market. Six things keep coming up.
1. Marketing loses the keys to its own channel
The people who own the message can't change it. A price. A disclaimer. A hero line. Every edit turns into a ticket, a prompt or a favour. Content velocity, the thing AI was meant to speed up, slows to the pace of whoever still understands each build. Ironic, right?
2. The cost moves from build to run
Vibe coding cut the cost of launch. It didn't cut the cost of owning the thing. Every bespoke property needs its own updates, hosting, fixes and onboarding for every new team member. Clients who budgeted for cheap builds are now paying for a pile of small maintenance contracts they never planned for.
3. Security becomes nobody's job
This one keeps me up at night. Veracode tested over 100 models and found 45% of AI-generated code samples failed security tests against the OWASP Top 10. Georgia Tech's Vibe Security Radar saw new AI-linked vulnerabilities jump from 6 a month in January 2026 to 35 in March (Cloud Security Alliance). On a fragmented estate there's no single place to patch, audit or monitor. Every site is its own risk, with its own owner. Or none.
4. The brand starts to drift
No shared components, no shared content model, and every property reads the brand a little differently. Type, tone, imagery, accessibility, cookie banners, legal copy. Small gaps stack up until the customer meets five slightly different companies.
5. Compliance turns into archaeology
Privacy updates, accessibility standards, regulatory copy. They all need to change everywhere at once. On a standardised system that's one change. On a vibe-coded estate it's a dig through repos, builders and logins to find where each line lives. Bring a shovel.
6. The client depends on people, not platforms
How each site works lives in one person's head. When that freelancer moves on, or the agency relationship ends, the client is left with something that works until it doesn't. Replatforming becomes the only way out, and round we go again.
Oh and here's the kicker. The next wave of AI, agents that read, reuse and personalise content, needs that content to be structured and consistent. Fragmented, hard-coded content is invisible to them. The shortcut taken to move fast with AI is now blocking clients from the AI that comes next.
Why agencies carry the risk too
It's tempting for agencies to see this as the client's problem. Or a nice cleanup opportunity. It's neither IMO. It hits the agency relationship first.
Trust erodes quietly. When a client can't edit their own site, they don't blame the model. They blame whoever built it.
Margin leaks into support. Unplanned fixes on one-off builds eat retainers and pull senior folk off creative work.
Handovers get harder. Inheriting someone else's vibe-coded project means reverse-engineering choices nobody wrote down.
Speed becomes a commodity. If anyone can prompt a site, a fast build isn't a reason to hire an agency any more. Lasting value is.
The agencies that come out of this well will be the ones offering clients something a prompt can't. A standard they can live with for years, not weeks.
What a standard looks like
Let me be clear. The answer isn't to stop using AI. AI is an amazing production tool and I love what it's doing for our industry. The answer is to give it rails, so every build lands in the same structure no matter who, or what, made it.
A standard worth having does five things:
One content model. Content is structured and kept separate from code, so marketing edits words, not components.
One editing experience. Every property, from the flagship site to the smallest microsite, is managed the same way.
Shared, maintained infrastructure. Hosting, security, updates and performance are handled once, centrally. Not site by site.
AI under control. Agents work inside the system through defined tools and skills. The agency picks the models and keeps token costs predictable.
The agency keeps the client. The standard powers delivery behind the scenes. The agency stays the trusted partner.
That's the thinking behind an Agency Web Ecosystem. We built Blutui inside a working agency to solve exactly this kind of fragmentation, long before AI turned the volume up on it. Developers keep full code-first freedom and the AI tools of their choice through MCP. Clients get one consistent, editable home for their content. Agencies get to move at AI speed without leaving AI spaghetti behind them.
If you want to go deeper on the cost of these gaps, have a read of our BluPaper, Compounding AI Chaos.
The bill is coming either way
Every industry that scaled fast without standards ended up paying to standardise. Web's no different. The only question is whether clients pay now, by choosing a shared foundation, or later, by unpicking dozens of one-off builds under pressure.
Agencies are best placed to lead that conversation. They know the brand. They hold the relationship. They see the mess forming first. The ones who bring clients a standard, rather than another fast build, will own the next decade of web work. I'd put money on it.
So if your clients are starting to ask how to change the headline, let's chat. Book a time here.
Talk soon. Gb
Sources
Cloud Security Alliance, Vibe Coding Security Debt: AI-Generated Vulnerabilities at Scale, April 2026 (citing Veracode, Georgia Tech Vibe Security Radar and Y Combinator data)
Blutui, Compounding AI Chaos
Author: Graeme Blake, Founder & CEO, Blutui
More Articles
-

Product 5 min read
How we built browser login for Courier with the OAuth device grantCourier now logs in through the browser with the OAuth device grant, and why we chose it over PKCE.
-

4 min read
Head in the Sand or Hair on FireMost agencies I talk to are standing somewhere on a line. At one end, nothing changes. At the other end, everything changes at once, with nobody holding the wheel.
-

Agency Life 6 min read
The Confident Obituary Con.UI Isn’t Dead. It’s Time to Double Down.
-

Agency Life 4 min read
Agency Advisory | 22 Oct 2026 (NZT)Legacy PHP Vulnerabilities and Your Clients' Websites