Can AI Build a Complete App or Website?

As of September 2026, this question has three annoyingly valid answers:

No.
Partially.
Yes.

Confused? Good. That means we can skip the usual AI marketing sermon and talk about what actually happens when you try to build something people are supposed to use.

AI can generate a website, an app, a WordPress plugin, or a browser extension. It can create hundreds of lines of code in seconds, explain them, change them, and occasionally break three unrelated things while fixing one button. What it cannot reliably do is turn a vague idea into a polished, secure, maintainable professional product without a competent human directing the work.

The important question is no longer whether AI can write code. It can. The real question is whether the finished result is any good.

You Still Need Real Skills

AI is clever, useful, and an enormous time-saver. It is good for fun and very good for work. Some models still suck at particular tasks, while others perform impressively. Benchmarks try to turn all of that into tidy numbers, but a coding test cannot reproduce every ugly real-world project with its conflicting requirements, old dependencies, strange server configuration, unpredictable users, and a client who remembers one essential feature two days before launch.

A real product is more than code that runs once on the developer’s computer. It includes the features, user interface, user experience, accessibility, error handling, security, compatibility, performance, installation, updates, backups, documentation, and all the unpleasant situations nobody mentioned at the beginning.

I have experience in network and computer support, web design, and development, and I still do not know everything I should know. I am not rich because I am apparently hopeless at selling my own skills, but let us stick to the topic.

Those skills matter because you need to recognize when AI is wrong. If you cannot tell good code from bad code, a solid solution from a fragile trick, or a sensible interface from a polished turd, you are not really developing a product. You are accepting whatever the machine happens to produce.

When Clients Expect One-Minute AI Results

I once had a client who wanted a plugin that showed shoppers how much more they needed to spend to qualify for free shipping. Existing plugins already did this, so why create another? Because some required overpriced subscriptions, while others lacked country-specific free-shipping thresholds, GeoIP detection, product-group exclusions, or compatibility with her shipping and multilingual setup.

To build it properly, I had to ask several questions about the requirements and edge cases. She became annoyed because AI was doing the coding, so apparently the whole thing should have taken one minute. I became fucking annoyed because AI can generate code quickly, but it cannot guess business rules the client has not told me.

After I finished the plugin and sent my invoice, she dropped me a note saying that my time was expensive and every new expense hurt her business. What the fuck was that supposed to mean? Her business expenses were real, but my working hours were apparently decorative.

AI accelerated the coding. It did not eliminate the requirements, compatibility work, testing, or responsibility. Yes, I could have produced something in one minute. It just would have had one minute’s worth of quality.

A Prompt Is Not a Product Specification

You can ask AI to make a WordPress plugin. I do that, and it works remarkably well. However, getting a genuinely good plugin involves more than typing, “Make me a nice plugin.” You need to define:

• What it should do and what it must never do.
• How each feature and user flow should work.
• What the interface should look like and who needs to use it.
• What should happen when something goes wrong.
• How data should be validated, stored, protected, and deleted.
• Which users have permission to perform which actions.
• How it should behave with caching, themes, other plugins, different PHP versions, and weird hosting environments.
• What happens during installation, updates, deactivation, and uninstallation.

And blah, blah, blah. Unfortunately, the blah, blah, blah is where much of professional development lives.

If you do not know what you want and have no relevant skills, AI may still make something. It might even look finished. That does not make it a professional product. Underneath the attractive buttons may be duplicated logic, insecure requests, missing permission checks, inaccessible controls, data loss waiting for an opportunity, or code held together by optimism and one deprecated function.

In other words, AI can make a shitty result very quickly. Progress!

AI Needs Direction, Correction, and Responsibility

AI needs precise instructions, but it also needs corrections. Sometimes it needs to hear a very clear, “No, this is not good in this situation.” You cannot treat every suggestion as brilliant merely because it arrived in two seconds and used confident vocabulary.

AI often chooses the most obvious implementation. That can be useful for a prototype, but the obvious implementation is not always the safest, cleanest, fastest, or easiest to maintain. It may solve exactly what you requested while ignoring what you meant.

Humans are extremely talented at writing incomplete requirements. AI is extremely talented at completing the wrong thing with confidence. Together, we are magnificent.

There is another detail the hype tends to skip: AI does not carry responsibility. It will not answer support emails, apologize for lost orders, restore a damaged database, or pay a client after a security breach. The person who publishes the code owns the consequences.

That is why you need to understand the project well enough to challenge the output. Ask why an approach was chosen, inspect the affected files, run tests, check edge cases, and sometimes throw the entire solution away.

AI Saves Time, but Good Work Still Takes Time

I no longer need to spend months coding certain projects. AI can reduce that work to weeks, and sometimes a small feature really can be completed in a day. That is a huge improvement. It is not magic.

I cannot publish a plugin that works beautifully in one artificial test and then falls apart on a real website. Real-world testing takes time. Tests reveal problems, problems require changes, and changes require more tests.

New ideas also appear during development. Some are useful. Others are complicated little bastards that add three settings, five edge cases, and almost no real value. “Nice to have” does not mean “worth having”. One of the developer’s jobs is knowing what not to build.

Every meaningful piece of AI-generated code should be reviewed. That can include manual review, automated tests, static analysis, security checks, or a second AI examining the first one’s work. Preferably, it means a combination of them. Different eyes notice different problems, even when some of those eyes live in a data center.

Professional Developers Use AI Too

Professional developers are not sitting in candlelit rooms, heroically typing every bracket by hand to protect the ancient honor of programming. They use AI for autocomplete, boilerplate, documentation, test generation, debugging, refactoring, explaining unfamiliar code, locating relevant files, and drafting migrations. More capable coding agents can inspect a repository, edit several files, run commands, and iterate on test failures.

But professionals do not become unnecessary simply because the typing is faster. Their work shifts toward architecture, requirements, review, integration, security, testing, and deciding whether the generated solution belongs in the product at all.

The 2025 Stack Overflow Developer Survey found that 84 percent of respondents were using or planning to use AI tools in development, and 51 percent of professional developers used them daily. Yet 46 percent of developers distrusted the accuracy of AI output, while only 33 percent trusted it.

That apparent contradiction makes perfect sense: developers use AI because it is useful, and they distrust it because they have actually used it.

Google’s 2025 DORA research described AI as an amplifier. It magnifies the strengths of effective organizations and the weaknesses of dysfunctional ones. The same is true for an individual developer.

AI can make a skilled person faster. It can also help an unskilled person create a much larger mess before lunch.

My Hi-Res Audio Player Died Young

I tried to make a quick Windows application, a real executable rather than another web toy. It was supposed to be a hi-fi player for Hi-Res audio files. The first version? Deleted. I gave up.

I will probably try again, but first I need to learn more about DACs, WASAPI, exclusive mode, drivers, audio formats, bit depth, sample rates, resampling, device switching, and what Windows quietly does to an audio stream when nobody is looking.

AI could produce a player interface and make sound come out of the speakers. That is not the same as producing a trustworthy hi-fi player. If the application claims bit-perfect playback, I need to know whether that claim is true.

I cannot simply ask the same AI that wrote the code whether its own work is excellent. Of course it thinks the answer is yes. It practically wrote the review before I finished asking.

That failed experiment was useful. It reminded me that AI can help me work inside an unfamiliar field, but it cannot instantly give me the judgment that comes from understanding that field.

Can AI Build a Complete Website?

Technically, yes. AI can create the HTML, CSS, and JavaScript, make the layout responsive, add forms, prepare metadata, and generate structured data in JSON-LD. It can also build on WordPress, a framework, or one of the growing number of AI website platforms.

For a simple brochure website, landing page, portfolio, or internal prototype, the result may be perfectly usable. I have moved several simple projects back to old-school HTML, CSS, and JavaScript because they do not need WordPress, a database, seventeen plugins, and a monthly emotional-support subscription.

AI is very good at helping with that sort of lean project. What fails to impress me is the visual result. An ordinary user may look at an AI-generated website and say, “Oh, that is lovely!”

I look at it and think, “It looks like a fucking template.”

The coding may be competent, but the design often has the same familiar ingredients: a large hero section, a gradient, several rounded cards, suspiciously generous empty space, generic icons, and a button announcing that we should “Get Started.”

Started with what? Apparently with looking like every software startup created during the same afternoon.

I then have to explain how the interface should actually look, which elements deserve attention, how the spacing should behave, what visual hierarchy I want, and which fashionable nonsense should be removed. One acceptable result took another three days and still looked slightly templated.

To be fair, that was partly my fault. “Make a nice website” is not a design system. I should have provided my own rules for typography, spacing, colors, components, breakpoints, interactions, and overall visual character.

That experience is a neat real-world demonstration of this entire article: vague instructions create generic results, then the human complains as though the machine ignored the brilliant design that never left his head.

And yes, the AI followed several good SEO practices and created structured data. Of course, I had to ask for them.

AI is helpful, but it is not psychic. At least not yet. Give the marketing department another six months and they will claim it is.

The Rise of the One-Day Coder

The downside of AI coding being available to almost everyone is that almost everyone can now publish software. App stores, plugin directories, extension libraries, package registries, and public code repositories will receive a great deal of what people politely call AI slop.

The problem is not that beginners are allowed to create things. Everyone starts somewhere, and AI can be a fantastic teacher. The problem begins when somebody generates an app in one day, performs almost no review, publishes it for strangers, and suddenly becomes a software company until next Tuesday.

A one-day coder may not provide long-term maintenance, answer support requests, patch security problems, follow changes in browsers or operating systems, or even remember which AI conversation produced the code.

Once the novelty disappears, users can be left with abandoned software that still handles their files, credentials, payments, or personal data. Lovely.

AI also makes it easier to create dependencies nobody understands. A person can publish code without knowing why it works, what could break it, or whether a copied package has introduced a vulnerability.

The application may function today, but maintenance begins tomorrow, and tomorrow is where many exciting new software companies mysteriously vanish.

Perhaps this flood will eventually teach people something useful: human judgment, accountability, maintenance, and support still have value in the AI age. Code is not finished merely because it exists.

So, What Is the Real Answer?

Can AI make an app or a complete website? Yes, especially if the project is small, the requirements are clear, and somebody competent reviews the result.

Partially, if AI does most of the implementation while a human handles the architecture, decisions, testing, design, security, and final responsibility. This is probably the most honest answer for serious work today.

No, if “make” means taking one vague sentence and autonomously delivering a polished, original, secure, accessible, maintainable product that survives real users and real problems without knowledgeable human involvement.

AI is improving quickly, and this article may age faster than I would like. Models gain new capabilities constantly. They can already handle larger projects and more complicated tasks than they could a short time ago.

However, even if the technology becomes dramatically better, someone still needs to decide exactly what the product should do, how it should behave in known and unknown situations, and whether the result is acceptable.

Now I have to get back to my latest plugin. I made major changes, which means I need to run more tests, find more problems, make more adjustments, and resist adding features nobody needs.

AI has made the work much faster. It has not made the work disappear. It takes time. It sucks.