How I Use AI to Build Websites Faster Without Cutting Corners

Some parts of my work now take a fraction of the time they used to. Others take exactly as long as they always did, and a few take longer, because there’s more to check. Knowing which is which is most of what it takes to do AI-assisted web development well.

I use AI at nearly every stage of building a website, from research to launch. My main tool is Claude Code, an AI coding agent that works directly in a project’s files: it reads the code, edits it and runs commands. On my own site, it wrote most of the WordPress theme. It didn’t decide what the theme should be, how the content should be structured, or whether the result was good enough to launch. That split is what AI-assisted web development looks like in my work: AI accelerates the workflow, and experience determines what should actually be built.

Here is how it works in practice, stage by stage, and what it means if you’re planning a website with someone who works this way.

The AI-Assisted Web Development Workflow at a Glance

The pattern in the right-hand column is consistent: AI produces, I choose and verify. The rest of this article explains why each row is split the way it is.

Research and Planning: Faster Answers to the Same Questions

A lot of web development is reading: a plugin’s documentation, an API’s limits, the code of a site I’m taking over. AI shortens this dramatically. It can read a whole codebase and explain how the pieces fit together, or summarize what changed between two versions of a library.

What it can’t do is decide which of those facts matter for this project. Planning is still where I spend unhurried time: what the site is for, who will edit it, what has to survive from the old site, and which content is shared across pages rather than written per page. When I rebuilt this site, the AI-generated prototype came with a handoff document that mapped every design section to a future template and listed what still needed real data. It turned out to be the most useful input of the whole project, because it turned a picture into a specification. I tell that story in I built a WordPress theme with AI.

UX Exploration and Prototyping: Options Are Cheap Now

Exploring three directions for a page used to mean days of work on two that would be thrown away. Now an option can take minutes, so it’s realistic to look at several before choosing, a shift I wrote about in how AI is changing the way we build websites.

That changes the conversation with a client for the better. Instead of debating a description, we can react to working versions. But more options make the choice harder, not easier, and the choice has to be made against the goals: who the page is for, what they need to believe, and what they should do next. A tool can make a dozen layouts. It has no reason to prefer one of them.

Development: The Agent Types, I Set the Rules

In development, AI does most of the typing on my projects. Converting a design into templates, wiring up fields, writing migration scripts and making the same change across forty files is exactly the kind of high-volume, well-defined work it does well.

The quality comes from the structure around it:

• Rules before code. Which files may change, which conventions to follow and what stays untouched. On my site, the design’s CSS was off limits, so any visual difference was a template bug rather than a styling debate.
• Small units. One section or one feature at a time, each small enough to check against the design on its own.
• Reading the diff. I read what changed, not the agent’s summary of what changed.

For WordPress projects in particular, I break this down stage by stage in my AI-powered WordPress workflow.

Debugging: Causes, Not Symptoms

AI is a fast debugging partner. It can trace an error through several files and propose a fix in seconds. The risk is that a plausible fix makes the symptom go away while the cause stays where it was.

A recent example from my own portfolio pages: cropped project screenshots looked slightly soft, even though the image files themselves were sharp. The cause was a detail of how WordPress builds responsive images. For lazy-loaded images it adds a sizes="auto" hint, which lets the browser pick a file as wide as the image’s box on the page. These images were cropped to fill their frames, so they were drawn wider than their box, and the browser kept choosing a file that was slightly too small. The fix was to tell the browser the width the image is really drawn at. Forcing the largest file everywhere would also have made the softness disappear, at the cost of every visitor downloading far more than they need.

That’s the kind of problem where experience earns its place: knowing which layer to look at, and which fix you’ll still be glad about a year from now.

Content and SEO: Help With Structure, Not Claims

AI is useful for content work in specific ways: outlining an article, checking that headings follow a logical order, finding pages with missing or duplicate descriptions, spotting broken links, and checking that code examples still work. When I reviewed this blog’s older articles, AI did the first pass on every post, flagging outdated code and weak structure, and every changed code sample had to pass an automated check before it went back in.

What it doesn’t do is make claims on my behalf. Anything published under my name, whether it’s a fact, a recommendation or an experience, has to be something I can stand behind. That rule matters more for search now, not less, as I explain in why an AI-generated website still needs good SEO, content and structure.

Testing and Launch: Checks I Can Run Again

The most valuable thing AI has added to my workflow isn’t code. It’s checks. An agent can write and run, in minutes, verification that would be too tedious to do thoroughly by hand:

• Syntax checks on every file.
• Every page rendered with debugging switched on, to catch warnings a browser would hide.
• Screenshots at desktop and phone widths, compared against the design.
• Scripted tests of the interactive parts: filters, search, forms, menus.
• A status check of every URL, including every redirect from the old site.

My rule is that I trust the checks, not the agent’s report, and that every check has to be capable of failing. A test that has never been seen to fail hasn’t proved much yet.

The Corners I Don’t Cut

Speed creates pressure to skip things. These are the parts of a project I keep for myself, however fast the rest goes:

• Architecture. How the site is structured, what it depends on and where its content lives.
• UX and design. Hierarchy, flows and every state, not only the happy path.
• Accessibility. Keyboard use, labels, contrast and zoom, tested rather than assumed.
• Performance. Measured on a phone, not judged on a fast laptop.
• Business requirements. What the site needs to do for the people who own it.
• Code quality and maintainability. Code that someone else could pick up next year.
• Production readiness. Tested on the environment it will actually run on, with backups and redirects in place.

None of these became easier because AI got better. They became a larger share of the work.

Does AI-Assisted Web Development Reduce Quality?

It can, if the time it saves goes straight into shipping sooner. It improves quality when that time goes into the things that used to get squeezed: more checks, more devices, more states, and more attention to the details that make a site feel finished. That’s the trade I try to make.

For a client, the practical effect is that working versions arrive earlier, so feedback happens on something real rather than on a description, and more of the project’s time goes into review. The bottleneck moves, too. Code is rarely what holds a project up now; decisions and real content are. A client who has their content, priorities and examples ready gets the most out of an AI-assisted process.

You can see the kind of work this produces in my portfolio, and read more about how I work on the about page.

Faster Is Only Useful in the Right Direction

AI has made me faster at nearly everything that can be clearly specified. It hasn’t made the specification any less important, and it hasn’t changed who is responsible for the result.

That’s the balance worth looking for in anyone building a website today: someone who uses these tools fully, and who can tell you exactly where they stop trusting them. AI accelerates the workflow. Experience determines what should actually be built.