How I built this website with AI: the learning curve

The website you are reading was built with AI. Not a template filled in by a chatbot, but a custom WordPress theme, a companion plugin and every page you see, built by me working alongside an AI assistant. I want to share how that actually went, because the honest version is more useful than the hype.

What we built

• A custom theme, Rashid, with its own design system: a clay accent colour, Fraunces for headings, IBM Plex Mono for text, uniform spacing and a dark mode.
• A companion plugin, Rashid Core, that holds the content types: Projects, Media & Talks, Events, Certificates, My Plugins and a Changelog, plus view counts, share buttons and security hardening. Keeping content in a plugin means it survives a theme change.
• Pages built with Elementor Pro on top of that theme, including sliders, working forms and a projects portfolio with real screenshots of client sites.

I built it on my own machine first (XAMPP on localhost), which let us break things safely.

How the work actually felt

Working with AI felt less like pressing a button and more like working with a very fast developer who needs clear direction. My messages were things like “remove the arrow on the buttons”, “make the spacing 30px”, “the image blends into the background in dark mode”. The AI did the work, checked it in the browser and reported back. I reviewed, pushed back and asked for changes.

The learning curve

• Be specific. Vague requests get generic results. The best outcomes came when I said exactly what I wanted and why, often with a screenshot.
• You still need to know WordPress. Deciding what belongs in a theme and what belongs in a plugin, how to deploy without breaking the site, how Elementor stores its data: the AI could do the work, but I needed to understand it to steer it and to judge the result.
• Test everything. The contact forms looked fine but rejected every submission because of empty field IDs. We only found it by actually submitting them. “It looks right” is not the same as “it works”.
• AI makes mistakes too. An automatic code formatter changed one lowercase value and caused PHP warnings on every page. A piece of database recovery advice had to be corrected. The fix each time was the same: verify, do not assume.
• Standards still matter. We ran the theme against the WordPress Coding Standards, reviewed security (escaping, nonces, file editing disabled, XML-RPC off) and audited every page on phone screens. AI made these checks fast, not optional.
• Taste and content are yours. AI can generate a hundred layouts, but it cannot decide what your site should say about you. Choosing what to show, and what to leave out, was my job.

What AI was great at

• Turning a reference design into a working theme quickly.
• Repetitive, careful work: creating 15 project entries with screenshots, or rewording content across the whole site.
• Checking its own work in a browser, on desktop and mobile.
• Explaining what went wrong when something broke.

What still needed me

• Decisions: design direction, what content to publish, what to keep private.
• Anything involving passwords or accounts. I never shared credentials; I signed in myself.
• Judging whether something was actually good, not just finished.

Would I do it again?

Yes. AI did not replace my skills; it multiplied them. Work that would have taken me weeks of evenings happened in focused sessions, and I spent my time on decisions instead of boilerplate. If you are a developer wondering whether to try it: start with a project you know well, stay in charge of the decisions, and test everything. That is the whole secret.