AI is rebuilding the internet. Nobody thought to fix the accessibility problem first.

Why AI-generated products are quietly deepening the accessibility gap, and where forward-thinking teams are getting ahead

Go look at the numbers. Around 96 to 97 percent of websites fail basic accessibility standards. One in five people globally live with a disability. Accessibility failures come down to recurring patterns: missing alt text, poor contrast, inaccessible navigation, broken keyboard interactions. However, these aren't edge cases from a decade ago. Right now, it's the norm for the overwhelming majority of what's on the internet.

And now we're rebuilding on top of all of it.

The speed gains from AI tooling have been unprecedented. Design and engineering teams are shipping more in a month than they used to in a quarter. But LLMs learned from the internet as it exists. They scraped it, ingested it, and now reproduce it at scale. When a landing page or UI gets generated in seconds, it's not being generated against the ideal, it's being generated against the average of everything that already exists. Which is, by most measures, completely inaccessible.

Before AI, you could at least argue the problems were isolated. A team ships an inaccessible feature. Another misses a key flow. The damage was real, but it stayed in one place. Now these same flawed patterns get reproduced across multiple products at once. Agents generate and deploy variations continuously. What used to be a contained problem becomes systemic. Fast.

At DevAlly, this is a pattern we have seen with customers across Fintech, HR tech, and EdTech. Teams aren't making a deliberate choice to deprioritise accessibility. They're moving quickly, adopting AI tooling, and realising six months later that accessibility was never part of how they were building. By the time it surfaces, design systems are already in use, components have been reused everywhere, and the issues are spread across core user journeys. You're not fixing a bug at that point. You're scoping a programme of work.

There are usually three moments that make it urgent. A deal slows down because a customer or procurement process raises accessibility requirements. A new market comes with compliance expectations nobody planned for. Or someone flags usability problems that can't be quietly closed and marked as low priority.

Standards like WCAG 2.1 AA (opens in a new tab), via EN 301 549 in Europe (EAA (opens in a new tab)), are showing up in procurement processes now as prerequisites. In the US, requirements linked to the Americans with Disabilities Act (opens in a new tab) or Section 508 (opens in a new tab) continue to shape how digital products get evaluated in education, public sector, and enterprise. These aren't distant considerations. It’s a global issue.

The velocity argument for AI is legitimate. Teams are genuinely building faster. What tends to get missed is that speed without the right foundations doesn't save time, it redirects it.

Engineering time that should go to new features starts going to fixing existing ones. Design systems need to be rebuilt rather than extended. The same accessibility gaps show up across multiple releases. The initial gain gets absorbed by rework, and nobody planned for it in the roadmap.

With DevAlly, product teams are not slowing down, but are making better decisions earlier.

• They are building accessibility into design systems from the start, with components that are accessible by default.
• Their prompts and templates include accessibility considerations, so AI outputs don't have to be corrected manually every time.
• They are leveraging automated testing in their CI/CD pipelines.
• They are sharing ownership of accessibility conformance across product, design, and engineering instead of one team chasing it as a late-stage task.

What this creates, in practice: smoother procurement, access to regulated markets, and less time spent going back over things that should have been right the first time.

There's a split happening in how teams are using this moment. Some are using AI to generate faster, on the same foundations they had before — scaling whatever was already there, including the gaps. Others are using the disruption as an opportunity to reset how their systems work.

What the second group understands is that if accessibility isn't part of the default, it doesn't get added later through goodwill. It gets added when there's a commercial reason, often under pressure, often at considerable cost. Whereas if it's built into the way products are created from the start, it compounds, with better outputs, increased uniformity, less risk and importantly, much broader market reach.

AI is rebuilding the internet. The question is what you embed in your foundation: accessibility standards, inclusive defaults, and the tooling to enforce them before the concrete sets.