Here’s the strategy underneath all of it: test on the site where a mistake costs nothing, before it ever gets near a site where it costs a client something. My own site goes first. Mel·lif·lu·ous, my music project, goes right behind it, small and forgiving enough to absorb whatever I’m not fully sure of yet. A client only sees a process after it’s already survived two rounds of me trying to break it.
Six months of discourse eating itself
I spent about six months watching the AI-and-web-dev discourse eat itself. Every other video: WordPress is dead, no-code is dead, hand-coding is back from the dead, usually with a thumbnail of someone’s face looking alarmed. The “right” answer changed every few weeks, which is an annoying thing for an industry to do to you while you’re trying to run a business in it.
Here’s what didn’t change: the fundamentals got more important, not less. Not despite the AI. Because of it.
That’s the whole thesis, actually. AI didn’t shrink the amount of judgment a site needs, it moved the judgment earlier, into the structure underneath, because now anyone can generate the surface. The version of me that wins the next five years isn’t the fastest to publish. It’s the one whose foundations are solid enough that new tools plug in without a rebuild.
What this round of updates actually was
So this round of updates wasn’t really about features. It was about infrastructure: a library of design references, a full icon system, a guidelines page that can’t drift out of sync with the live site, a cleanup pass on how links behave sitewide. None of that is what a client site needs on day one. All of it is what decides whether a site is still easy to build on two years from now, or whether every future change means picking through something nobody documented.
Novamira, and finally leaving Elementor
Novamira is the tool that made most of this possible. It lets AI reach directly into a site’s back end and check what’s actually true instead of guessing, then make the change instead of just suggesting it. That access is also why I finally broke up with Elementor after years of using it. (My heart aches a little.) Not because it stopped working. Because it had spent years layering systems on top of systems, and I could feel it stop keeping up.
What’s in its place
In its place: a naming convention I can carry across every project, not just this one. A workflow that doesn’t care which tool I’m in, Claude, VS Code, Antigravity, so switching between them doesn’t cost me my place. A design system actually wired to the live site instead of sitting next to it: one source for colors and type, and when it changes, the site changes with it. That’s the part that never shows up in a screenshot. A site can look finished and still be built on nothing underneath it. This one isn’t.
The compounding effect
Here’s where it compounds. I’ve got several sites running at once now, and I’ve stopped treating them as separate projects that happen to share a calendar. Mel’s site is no longer built off mine, I’m building off Mel’s cleaner version. A client project got built from nothing, specifically to test whether the process held up with no old work to lean on. A mistake caught on one gets checked for on the next before it gets the chance to repeat. I’m not starting from zero each time. I’m building the process, and the sites are how I test it.
A lot of what’s available off the shelf right now, ready-made setups, AI-generated shortcuts, gave me ideas to react to and not much else. What I ended up with is specific to how I actually work. Some of it translates. Figuring out which pieces are worth keeping is its own job, on top of everything else.
What actually changed, and what didn’t
Which is the actual point. Building something meant to last is still a long process, still section by section, decision by decision, nothing skipped. What changed is I have real access now instead of suggestions I have to type in myself. What didn’t change is the checking, the testing, catching the thing that looks fine and isn’t. The tool got faster. The work didn’t get shorter. It just got more of what it always needed: someone paying attention the whole way through.
Most of the industry right now is optimized for how fast a site can go up. I’m optimized for whether it’s still standing, still working, and still finding you the right people two years in. That’s the harder problem to solve for. It’s also the only one that matters once a site is live, which is why I run my own fleet of guinea pigs before anything gets near yours.
