My studio sells premium WordPress themes, and I am not an optimist about premium WordPress themes. The reason is AI. Both halves of that sentence are true, which is exactly why I do not trust the obituary everyone is racing to publish. I have watched what actually happens when a real project tries to leave, and the grass on the other side is greenest in the screenshots.
- Path A
- vibe code it yourself
- Path B
- an AI site builder
- Path C
- the hybrid
The one thing AI genuinely fixed
Buying a theme used to carry an unadvertised cost: onboarding. Days, sometimes weeks, spent learning somebody else’s product. Documentation that reads like a lazily written label on the back of the box. A page builder with a philosophy. Ten small breakages and three support tickets, and only then the change you actually wanted, a logo swap and a new CTA in the footer.
AI collapsed most of that, and deserves the credit. An assistant reads the theme, finds the setting, writes the override. That part of the old world is not coming back, and good riddance. The question is what you walk into instead, and there are two doors on offer.
Path A: vibe code everything
The pitch is real: a $20 assistant subscription, an evening of prompts, and your site is live on Vercel or GitHub Pages before midnight. It works, in the sense that something appears at a URL.
Then the details surface. GitGuardian scanned public repositories for its 2026 report and found AI-assisted commits leak secrets at a 3.2% rate, roughly double the 1.5% baseline, inside a year that added 28.65 million newly exposed secrets. When a researcher scanned 1,645 apps built with Lovable, a popular vibe-coding platform, 170 of them, about one in ten, had databases readable by anyone on the internet. That one became CVE-2025-48757. Credentials sitting in page markup stopped being a joke and became a measurable category.
Past the security there is the sameness. The same layout, the same components, the same design the model reaches for every time, assembled with the confidence of a kindergarten craft project. The person from the reels promising agency-grade design out of ten random skills got paid the moment you subscribed. That was the product.
Path B: an AI site builder
The builders are more honest, and the good ones are genuinely good at what they cover. Two problems remain, one small and one structural.
The small one is the catalog. Wix advertises 2,000-plus templates, and the Framer marketplace sits in the same range by third-party counts. The free WordPress directory alone holds 8,563 themes, with ThemeForest adding five figures on top of that. For yet another B2B SaaS that is definitely not like the others, a template exists. For a catfish fishing charter, or magic mushroom cultivation in a jurisdiction where that is a website you can legally have, the long tail is still a WordPress feature.
The structural one is what sits behind the frontend, because most AI builders are frontend builders. Framer ships a real CMS, with a cap of 1,000 collection items on the Pro plan, a number worth comparing against your archive before you sign. v0 and Bolt do not pretend otherwise and hand the backend to Supabase. Everyone talks about Supabase. Almost nobody has read the free tier: 500 MB of database, two active projects, and a project that pauses itself after one week of inactivity. Your client’s site, paused for inactivity. Picture the phone call.
The realistic output of Path B without a paid stack underneath is another todo app or a six-slide portfolio. Fine things. Not a business.
The two weeks do not disappear, they relocate
Here is the accounting nobody puts in the demo video. The old two weeks went into reading theme documentation and arguing with a page builder. The new two weeks go somewhere else entirely:
- Bolting on some kind of working admin area, because content lives somewhere and somebody has to edit it.
- Writing legal pages from scratch, because the generator did not think about cookies, refunds, or your jurisdiction.
- Closing the security holes the assistant shipped, starting with whatever it left in the client-side code.
- Moving off the free container onto a server you would trust with a client invoice. The Vercel Hobby plan is non-commercial, per the plan documentation.
The bill did not shrink. It moved. And unlike the old bill, this one arrives after launch, when you are least in the mood to pay it.
Path C: the hybrid, which is where I landed
So who still buys a WordPress theme in 2026? People who have met both doors above and want the boring parts already solved. My answer as a builder is a hybrid, and it is the only reason my studio is still in this niche.
WordPress stays as the container, because a mature ecosystem of admin, roles, auth, content modeling and hosting is not something you rebuild by Friday. The theme carries what it should have carried all along: the design and the functionality of your vertical, nothing else. Underneath, instead of a dozen unrelated plugins, one framework: our JKD Toolkit runs the settings, the fields, and the moving parts of the theme. And around it a hub with the unglamorous essentials, licenses, updates, documentation and support in one place, with nothing bolted on for the screenshot.
The bill did not shrink. It moved, and it arrives after launch.
The lazy theme is dead, and it earned it
The ordinary premium theme, the one with a gorgeous cover, a lazy author and nothing behind it, is genuinely gone, and the Reddit threads mourning it are mourning a business model, not a product anyone loved. The niche around it is smaller every quarter: WordPress is down to 40.7% of the web, and even granting that part of this year’s drop may be a change in how W3Techs counts, the direction holds; marketplace unit sales have fallen by roughly two thirds since 2018, per a WPBay compilation of published Envato stats. I wrote about how the platform got here, and none of it was sudden.
What remains of the niche belongs to themes that ship as systems: container, design, engine, control panel. That is the whole bet. For the first time in a while, it feels like a good one.
Last updated
