The visibility problem agencies haven't noticed yet

Across my career I've worked in agencies that operated on fundamentally different philosophies about what they were selling. In one model, you start from the best solution, the architecturally sound, performant, accessible version of the thing, and negotiate down based on budget, timeline, and how much the client is willing to prioritise quality. The client gets a recommendation. They make an informed decision. The agency's reputation lives or dies by the quality of the work it puts into the world.

In the other model, you start from the simplest, cheapest version and add complexity only if the client asks and pays for it. Quality isn't the default. It's an upgrade. The agency's incentive is to ship, invoice, and move on.

Both models have coexisted in this industry for as long as I can remember. And for most of that time, the second one worked, not because it produced better outcomes, but because clients had no reliable way to evaluate what they were actually getting. They bought on relationships, on delivery speed, on the confidence of the person in the room. The gap between what was promised and what was built was invisible. And invisible problems don't cost you accounts.

That gap is not invisible anymore.

The audit tools have existed for years. Lighthouse, WAVE, PageSpeed Insights, all free, all browser-accessible, none of them requiring a developer to use. What they did require, until recently, was someone who could interpret the output. A Lighthouse report returns scores, diagnostics, and recommendations written in technical language. A failing accessibility audit returns a list of WCAG violations that mean nothing to someone who has never heard of WCAG. The information was always there. The fluency to act on it wasn't.

That barrier is gone. Paste a Lighthouse report into Claude, or ask Claude to fetch your URL directly, and you get a clear, plain-English breakdown of what's wrong, how serious each issue is, and roughly what it would take to fix it. No developer required. No agency mediation required. A marketing manager, a product owner, a CTO who has never written a line of code, any of them can now get a detailed quality breakdown of a website in under ten minutes, for free, with enough context to ask hard questions.

The tools didn't change. The ability to read them did.

Here's what those tools are about to reveal at scale. The WebAIM Million report, an annual audit of the top one million websites, found that 95.9% of home pages have detectable WCAG failures in 2026, up from 94.8% the previous year. The average page now has 56.1 distinct accessibility errors, a 10.1% increase year on year. WebAIM attributes this rise directly to increased reliance on third-party frameworks and what they call "automated or AI-assisted coding practices." The tools that were supposed to accelerate quality are, in aggregate, degrading it.

These numbers exist because of how most web work has been commissioned and built. Accessibility wasn't in scope. Performance budgets weren't enforced. SEO requirements were satisfied at a surface level and never stress-tested. In the quality-first model, those things get negotiated in. In the commodity model, they get left out.

For years, leaving them out had no visible consequence. The client didn't run the audit. They looked at the design and it looked fine. The fragility underneath was somebody else's problem to discover later, usually the client's, usually expensively. Tricentis's 2026 Quality Transformation Report, drawing on a survey of over 2,500 senior technology leaders globally, puts the cost of poor software quality at over a million dollars annually for one in five organisations. The same report found that 60% of organisations are now knowingly shipping untested code, driven by leadership pressure to prioritise speed over quality. Gartner projects that prompt-to-app approaches will increase software defects by 2,500% by 2028. The trajectory is already visible in the data. The industry is building faster and more carelessly at the exact moment clients are gaining the tools to see it.

I've written about what this looks like from the engineering side, the technical debt accumulating inside codebases, the review layers buckling under AI-generated volume, in What AI Can't Compress. The question here is different: what happens when the client can see it too?

There's something important about the timing of this shift. Clients know agencies are using AI. They've read the same articles. They've seen their own teams start using it. They understand, at least in outline, that development is faster and cheaper than it was two years ago.

So when a project comes in slow, inaccessible, technically fragile, and over budget, the conversation has changed. It used to be: they didn't have enough time, it's complex, these things happen. Now it's: they had the tools and still didn't care.

That reframing matters enormously. The old failure mode was a delivery problem. The new one is a values problem. Delivery problems are recoverable. You add a sprint, you discount an invoice, you move on. A client who concludes that quality was never the priority doesn't want a fix. They want a different agency.

The agencies most exposed by this aren't necessarily the ones cutting corners because of AI. They're the ones that were cutting corners before AI and are now doing it faster. AI hasn't introduced new failure modes to the industry. It has accelerated the existing ones and handed clients a torch to find them with.

The commodity model, at its core, is a bet on invisibility. Start cheap, add on if asked, ship it, move on. That bet made sense when clients couldn't see what they were getting. It makes far less sense when a non-technical stakeholder can paste a URL into an AI assistant and receive a plain-English report on why the site fails basic quality standards.

The agencies that built their model on technical opacity, on being the only party in the room who could assess what had been built, are running out of the conditions that protected them. The argument "trust us, this is complex" holds less weight when the client already knows what a Lighthouse score means.

What survives is the quality-first model: agencies where the conversation with the client starts from what's right and negotiates down from there, rather than starting from cheap and hoping nobody looks too closely. Not because quality is morally superior, though it is, but because it's the only model that holds up when the client can audit the work themselves.

The differentiator was never speed. Agencies that competed on speed were always one cheaper competitor or one faster tool away from losing the argument. It was always judgment: knowing what to build, why it matters, and how to have an honest conversation with a client about the gap between what they want to pay for and what their users actually need. That judgment doesn't sit in a Lighthouse score. It lives in the decisions made before the first line of code is written, in the architecture that determines whether the output is sound, and in the willingness to push back when the brief would produce the wrong thing. As I argued in What AI Can't Compress, judgment is the one thing AI cannot compress. What's changed is that previously it was bundled inside a layer of technical complexity that made it hard to separate from the rest of the work. Now that the complexity layer is legible to anyone with a browser and an AI assistant, the judgment has to stand on its own.

What makes this moment genuinely new is who raised the bar. Technical standards have always improved over time, but the pressure historically came from within the industry, framework evolution, browser capability, shifting best practices. The people doing the work decided when good enough moved.

That has changed. The bar for acceptable quality is now being raised by the people paying the invoices, because for the first time they have the means to measure it. The legal pressure is real too: the European Accessibility Act came into force on 28 June 2025, across all 27 EU member states. French disability organisations filed lawsuits against major retailers within days of the deadline. Regulators across Europe are actively monitoring. The website an agency built isn't just a design asset anymore. It's a liability if it fails standards that any free tool can now verify in minutes.

Most agencies haven't adjusted to this. The ones still operating on the commodity model are exposed, whether or not they know it yet. The ones operating on the quality-first model have always had something worth paying for. The difference now is that the market is starting to be able to tell which is which.

The audit is free. The excuses are running out.