# AI-generated code still needs an adult in the room > AI writes code fast. Production WordPress sites still need judgment, testing, security review, and someone who owns the system. Source: https://larsik.com/blog/ai-generated-code-still-needs-an-adult-in-the-room/ Business # AI-generated code still needs an adult in the room AI writes code fast. Production WordPress sites still need judgment, testing, security review, and someone who owns the system. July 5, 2026· Updated August 4, 2026· 5 min read·[Lars Koudal](/about/) - AI - software quality - WordPress - security - technical review ![A calm senior developer reviewing AI-generated code before deployment](/images/blog/ai-generated-code-still-needs-an-adult-in-the-room-w400.webp) AI can write code. That part is no longer interesting. The more important question is whether the code should be used, where it should be used, what it might break, how it should be tested, and who is responsible when it goes wrong. That is the part a lot of people are trying to skip. I use AI. I like AI. I think it is one of the most useful tools we have had in years. But I do not treat AI output as finished work. I treat it as a draft, a suggestion, a fast assistant, sometimes a useful second pair of eyes. Not as the person in charge. ## The dangerous part is confidence The problem with AI-generated code is not that it is always bad. Often it is surprisingly good. Sometimes it is cleaner than what a rushed developer would have written. Sometimes it gets you 70 percent of the way in a few minutes. The dangerous part is that it often looks finished. It can have good formatting, sensible variable names, comments, familiar patterns and an explanation that sounds confident. If you are not used to reading code, that confidence is persuasive. But confident output is not the same as correct output. Stack Overflow’s 2025 Developer Survey found that more developers actively distrust the accuracy of AI tools than trust it. Only a small percentage said they highly trust AI output. That is not because developers hate new tools. It is because developers know the cost of being almost right. Almost right code can pass a quick visual check. Almost right code can work on your machine. Almost right code can break checkout, expose private data, slow down a site, create SEO problems, bypass a cache rule, duplicate tracking, conflict with a plugin or fail two months later when something else updates. ## WordPress makes this even more important WordPress is flexible. That is why it is used everywhere. But flexibility also means that a website can become a layered stack of theme code, plugins, custom snippets, page builder logic, hosting rules, caching, third-party scripts, forms, WooCommerce, payment providers, SEO plugins, security plugins and a few old fixes nobody remembers. An AI tool does not automatically understand that history. It can write a function. It can suggest a hook. It can produce CSS. It can generate a small plugin. It can explain a SQL query. But it does not know the business consequence of breaking the contact form, the checkout flow or the technical SEO setup unless someone gives it the right context and checks the result. That is where the adult in the room matters. ## What “adult in the room” actually means This is not about age. It is not about being anti-AI. It is about accountability. Someone has to ask the boring questions: - Does this solve the actual problem, or only the visible symptom? - Is it safe for production? - What happens if the plugin updates? - What happens if the theme changes? - Is there a rollback path? - Does it affect performance, SEO, security, forms, payments or analytics? - Is this code needed at all? - Could a setting, a smaller fix or removing something be better? AI is good at producing options. Experience is good at filtering them. ## Speed without control becomes expensive Google’s DORA 2025 report describes the practical trade-off clearly: AI can accelerate software development, but without robust control systems, testing, mature version control and fast feedback loops, increased change volume can expose weaknesses downstream. That sentence should be printed above every “just let AI fix it” project. More code is not automatically progress. More changes are not automatically improvement. More output is not automatically value. A small website can suffer from the same pattern as a large software team: too many changes, too little review, unclear ownership and no one looking at the system as a whole. ## The new value is judgment For years, the expensive part of software was production. Getting something built took time. AI changes that. It lowers the cost of a first draft. It makes experimentation faster. It can help small teams move with less friction. That is good. But when production becomes cheap, judgment becomes more valuable. The person who can say “this is fine” is useful. The person who can say “this will break later” is more useful. The person who can say “do not build this at all” might save you the most money. ## How I use AI in real work I use AI as a tool, not as a replacement for responsibility. It can help with research, drafts, refactoring suggestions, test cases, documentation, quick prototypes and alternative approaches. It is especially useful when I already know the context and can judge the answer. But I still want to know what changed. I still want to test. I still want a rollback plan. I still want to understand the trade-off. That is true whether I am working on WordPress maintenance, a WooCommerce issue, a security cleanup, a performance problem, SEO tooling or a custom integration. AI can speed up the work. It should not remove the review. ## This is the real consulting opportunity A lot of businesses do not need more code. They need someone who can look at their website, their plugins, their hosting, their SEO setup, their tracking, their security and their maintenance process and say what actually matters. Not what is fashionable. Not what a tool flagged. Not what a random AI answer suggested. What matters. That is why I think experienced technical consulting becomes more important in the AI era, not less. The web is about to get flooded with more generated content, more generated code, more generated fixes and more generated confidence. Someone still has to decide what should go live. AI can write the code. Experience decides whether it belongs on your website. ## Related on Larsik - [About Lars Koudal](https://larsik.com/about/) - [30 år med kode: hvad erfaring betyder](https://larsik.com/blog/30-aar-med-kode-hvad-erfaring-betyder/) - [WordPress maintenance](https://larsik.com/maintenance/) - [Technical SEO](https://larsik.com/services/technical-seo/) - [PageSpeed and performance](https://larsik.com/services/pagespeed-performance/) - [WP Security Ninja](https://wpsecurityninja.com/) ## Sources and further reading - [Stack Overflow Developer Survey 2025: AI accuracy and trust](https://survey.stackoverflow.co/2025/ai) - [Google Cloud: Announcing the 2025 DORA Report](https://cloud.google.com/blog/products/ai-machine-learning/announcing-the-2025-dora-report) - [GitHub Octoverse 2025](https://octoverse.github.com/) - [Patchstack: State of WordPress Security in 2026](https://patchstack.com/whitepaper/state-of-wordpress-security-in-2026/) - [Larsik: About Lars Koudal](https://larsik.com/about/) ## Frequently asked questions Can I put AI-generated code straight into production WordPress?+ You can, but you should not without review. AI is fast at producing plausible code; production needs testing, security judgment, and someone who understands the rest of the site. What should a human review before deploying AI code?+ Security (escaping, capability checks, secrets), performance, conflict with existing plugins/themes, whether the change belongs in a child theme or custom plugin, and a rollback plan. Do you help review or fix AI-generated WordPress changes?+ Yes. That is a common engagement: stabilize what broke, clean up brittle snippets, and leave the site maintainable — see WordPress & WooCommerce services or contact.