Image of man with hands by face, he codes by hand

Why your website’s core should stay human-built.

Wednesday 08th July 2026

AI is changing how digital work gets made. It can remove friction, accelerate production and help teams move with more pace than they could on their own. But pace is not the same as precision, and output is not the same as judgement.

That distinction matters most in the parts of a website people rarely see, but always feel. The core of a platform determines how content behaves, how components scale, how accessible the experience is, and how gracefully the whole thing copes with change. When that layer is sound, everything above it works harder. When it is not, even the best-looking site is built on compromise.

That is why Foundation is built the way it is. Fiora’s Drupal-based design system is designed to stay lean, fast and accessible, while giving teams the flexibility to create polished business websites without dragging in unnecessary complexity or bloat. In that context, keeping the core human-led is not a romantic attachment to old ways of working. It is a disciplined choice about quality.

The risk is not mess. It is confidence.

Much of the case for AI-generated code rests on speed. If it can produce more, faster, the assumption is that teams gain efficiency and little is lost. The reality is less tidy.

CodeRabbit’s analysis of 470 real-world open-source pull requests found that AI-generated code introduced 1.7 times more issues than human-written code, with AI-authored changes averaging 10.83 issues per pull request compared with 6.45 for human-only pull requests. The defects were not cosmetic. They appeared across the categories that make systems hold together or fall apart: logic, security, maintainability and performance.

That is the real tension. AI does not usually fail noisily. It fails convincingly. It produces code that looks plausible, reads cleanly and gives the impression of completeness, while hiding problems in the places that matter most. For a core system, that kind of confidence is far more dangerous than obvious rough edges.

Foundations need authorship

A website platform is not just code. It is a set of decisions that will be repeated again and again, across templates, modules, components, content models and future builds.

Those decisions need context. Human developers can see the trade-off behind the implementation, not just the implementation itself. They can recognise when a requirement needs interpreting rather than translating, when an exception matters more than the rule, and when a technically valid answer is strategically the wrong one. That is what authorship brings to a system: responsibility, intent and the ability to think beyond the immediate output.

This is exactly why a reusable platform should be built with care. CodeRabbit’s findings suggest that AI-generated code remains weakest where systems need the strongest judgement, especially in logic and correctness. For a one-off prototype, that may be tolerable. For the architecture that underpins multiple business websites, it is not.

Security, performance and accessibility are part of the craft

The same logic applies to security, performance and accessibility. All three are easy to undermine and expensive to repair.

CodeRabbit’s findings show that AI-generated code carries a heavier review burden because the defects are more frequent and often sit in higher-risk areas. That matters because the cost of a weak foundation is rarely obvious at launch. It shows up later, in slower pages, heavier maintenance, brittle integrations and preventable risk.

Accessibility deserves equal weight here. Drupal’s own guidance stresses semantic HTML, careful use of ARIA, and public-facing code that strives to meet WCAG 2.2 AA. That makes it a strong base, but only if those principles are carried through the design system itself.

For Foundation, accessibility cannot be a finishing pass. It has to live in the patterns: in headings, forms, navigation, interactions, templates and markup choices reused across every build. That is another reason the core should stay human-led. Accessibility is not just a checklist. It is a discipline of clarity and consistency, and those qualities are far easier to protect when the system is intentionally authored rather than automatically assembled.

AI still belongs in the workflow

None of this is an argument for refusing AI. Used properly, it can be valuable. It can help teams move through lower-risk tasks more efficiently, support documentation, speed up repetitive work and create momentum around delivery.

But a useful assistant is not the same thing as a reliable architect. For a platform like Foundation, the sensible line is clear: let AI support the process, but do not let it define the system. Keep the architecture human-led, keep the standards intentional, and keep responsibility with the people who understand the consequences of the code they ship.

Seen that way, the real story is not what AI replaces. It is what it leaves behind. That is where people still matter most.

AI reveals where people matter most

The interesting thing about AI is not that it replaces human creativity. It is that it reveals where human creativity matters most.

AI is excellent at removing drag. It can help teams explore faster, test ideas sooner, document more clearly and move through the repetitive work that used to slow everything down. Used well, it creates more room for the work that actually deserves human attention.

That is exactly why the human-led core of Foundation matters. The value is not in any one person having the answer. It is in the push and pull between different kinds of thinking: the developer considering structure, the designer shaping experience, the strategist testing meaning, the content lead asking what is clear enough, the accessibility lens asking who might be excluded, and the client team grounding it all in what has to work in the real world.

That tension is not inefficiency. It is quality control with a pulse.

Foundation makes sense because it does not ask AI to do less. It asks people to use AI better. Let AI accelerate the search, widen the options and reduce the admin, but keep the judgement, the taste, the responsibility and the architecture with the people who understand what the system is meant to become.

What clients are buying into

Clients are not buying code for its own sake. They are buying confidence: that the platform will perform, that it will scale, that it will remain usable and maintainable, and that it will not create hidden problems in the name of speed.

A human‑led core gives teams a clearer view of the risk they are carrying than an AI‑generated one. CodeRabbit’s data shows that AI‑written changes arrive with more issues on the table, which means more to review, more to fix and more that can slip through unnoticed if the system itself isn’t designed with care. For a business website, that difference is not academic. It affects resilience, review burden and long-term cost.

It also supports the things clients increasingly expect to be there from the start, not fixed later. Drupal’s accessibility standards focus on readable markup, careful use of assistive roles and a clear line of sight to WCAG 2.2 AA. In practice, that encourages teams to treat accessibility as part of the platform’s design language, not as a check‑box at the end of the build.

The result is a more sensible relationship with AI. Use it to help the work move faster, but do not let it take control of the foundations. The future of good digital work is not human versus machine. It is humans with better machines, building things that still feel as if someone cared.

Sources

foundation

code